BlueDot Rapid Grants↗

Applications: Open
Apply ↗All 795 grants ↓
Programme of BlueDot Impact·Founded 2022 (began in 2021 as reading groups in Cambridge)·San Francisco (headquarters moved there in 2026); staff also in London
“Funding of up to $20,000 for time and resources to make progress on AI safety or biosecurity - from exploring a promising direction to carrying out a concrete project.”In its own words: bluedot.org
TechnicalGovernanceField-buildingCompute

BlueDot Impact's fast small-grants programme: up to $20k for individuals to explore an idea, do research or build something in AI safety or biosecurity. An updated version launched on 25 Nov 2025; an April EA Forum post reported 77 rapid grants totalling over $50,000.

Show detailsHide details

Small fast grants for individuals working on AI safety or biosecurity, covering research and learning, projects and experiments, events and community, and travel. Applicants send a short plan and budget, and BlueDot reviews it and emails a decision.

Typical grant $2k ($1k–$4k)

On applying “We look for a worthwhile direction, evidence you can make progress, and a clear role for funding.” bluedot.org

Itemised here $3.3M in 795 grants

’25
’26
Its grants page ↗Checked 29 Oct 2026

Money flows

Where its money came from and where it went. Pick years and what to show; hover a box or ribbon for details, click one to focus it, click the background to see everything.

How to apply

Applications
Open
Rolling
Who can apply
Individuals
Where
BlueDot can’t pay grantees in some countries, including China, India, Russia and Iran.

Open application, about 15 minutes, with a plan and budget: https://bluedot.org/grants/rapid. Funds can't be sent to some countries, including China, India, Russia and Iran.

Biggest recipients

All grants

795 grants totalling $3.3M, newest first. Recipients link to their own pages; people who received grants without a public profile are shown by number.

2026 780 grants · $3.3M
DateRecipientAmountPurpose
Oct 2026Individual #7662$20kLaunching Convergent Robotics' first safe-physical-AI experiments
Oct 2026Individual #8945$17kRunning Plures's first Brazilian AI safety essay contest
Oct 2026Individual #3329$10kTesting progressive misalignment from recursive self-distillation
Oct 2026Individual #7473$9kEvents RFE: UChicago AI Safety / XLab - AI in recent events panel discussion
Oct 2026Individual #9870$7kMentoring six economists in a SPAR project estimating the economic value of frontier AI capability gains: who
Oct 2026Individual #8894$7kDynamics and geometry of attractor states in long LLM-LLM interactions
Oct 2026Individual #2361$3kJHU campus-wide briefing on the HF Incident and on AI-Bio risks
Oct 2026Individual #4008$2kHosting The AI Girls Club's Lightning Round in London
Oct 2026Individual #5935$2kEvents RFE: Helsinki - AI Safety Now: What's Happening, and What Can We Do?
Oct 2026Individual #7033$2kRunning an economics-of-recursive-self-improvement lunch series
Oct 2026Individual #3223$2kAttending EA Global New York to choose an AI safety organization path
Oct 2026Individual #9721$1kWeight Oracles: LLMs that detect backdoors in neural networks by reading raw weights, zero-shot. Spotlight pap
Oct 2026AI Safety Barcelona$1kEvents RFE: AI Safety Barcelona source
Oct 2026Individual #9223$1kAttending EA Global New York to test an AGI economics institute
Oct 2026Individual #8929$1kEvents RFE: Rovereto - Public event in Rovereto (Italy) on 20/10/2026 on AI Safety and specifically on Loss of
Oct 2026Individual #2488$1kEvents RFE: Rice University — Understanding Recent AI Agent Incidents and Catastrophic Risk.
Oct 2026Individual #5607$1kAttending EA Global New York to test a frontier-AI auditing organization
Oct 2026Individual #6952$1kHosting a technical AI safety career event in Tel Aviv
Oct 2026Individual #1746$900West-China AI technical governance cooperation (developing a shared expectations for model AI-bio behavioral e
Oct 2026Individual #2275$900Cape Town: 'The AI Doc' screening and facilitated discussion
Oct 2026Matilda$775Attending COLM to develop autonomous-agent evaluations source
Oct 2026Individual #4037$700Events RFE: Sydney University
Oct 2026Individual #5034$650Hosting a catastrophic-AI-risk talk and discussion in Tokyo
Oct 2026Individual #5052$600Events RFE: Vancouver Unitarian AI Safety Event
Oct 2026Individual #9457$550Hosting a Barcelona community event on catastrophic AI risk
Oct 2026Individual #4522$500Moving from nuclear security into AGI policy and political strategy. Right now: leading the Yale Existential R
Oct 2026Individual #7319$500Ghent: Powerful AI is coming. What can you do to make sure we get it right?
Oct 2026Individual #9488$500Field building and event organising; foundational alignment and interpretability research; learning and shapin
Oct 2026Individual #9147$400Understanding catastrophic AI risk for social activists
Oct 2026Individual #6941$400The AI Doc: Or How I Became an Apocaloptimist Film Screening + Panel Discussion
Oct 2026Individual #5328$200The Hugging Face incident: what it means for your agents and products. For women building with AI.
Oct 2026Individual #8177$150Testing how AI-agent self-conceptions affect multi-agent misalignment
Sep 2026Individual #6973$20kEmergency AI Safety Comms Fellowship for Undergrads
Sep 2026Individual #2834$18kExpanding the London Interface for AI Safety
Sep 2026Individual #2503$15kARENA-based technical AI safety programme in Berlin (founder living costs)
Sep 2026Individual #5586$11kTesting diversity audits for AI-agent swarms
Sep 2026Individual #6637 (with Individual #2254, Michał Burzyński, & Shomit Basu)$11kA new LLM red teaming technique using RL and investigator agents. Current results show that we beat or match t source
Sep 2026Individual #4026$10kConvening a mathematics-for-AI-safety workshop in Bonn
Sep 2026Unknown Recipients$10kTesting human accountability in multi-agent AI swarms source
Sep 2026Individual #8044$10kMERe AI '26: a two-day workshop at Toronto's Vector Institute where researchers build a shared taxonomy of mor
Sep 2026Individual #8478$10kI want to make pretraining safety techniques cheaper by using them during distillation. I am doing this as a c
Sep 2026Individual #2713$10kRL reward hacking mitigation by controlling learning and generalization
Sep 2026Individual #9543$9kSupporting three months of AI safety research on scheming and collusion
Sep 2026Individual #3175$9kTesting the founding team and strategy for a new AI safety research organization
Sep 2026Individual #8788$8kTesting robust inoculation prompting under reinforcement learning
Sep 2026Individual #5502$7kScalable interpretability research, extending Activation Oracles. Training models to report on their own inter
Sep 2026Individual #1852$7kAssistants rewarded for the human's long-term power, not an inferred goal (Heitzig & Potham 2025): convergence
Sep 2026Individual #7207$7kAI Safety track at Investigathón (research hackathon) 2026 (UBA, ~250 undergrads): 6-month research scholarshi
Sep 2026Individual #6864$6kWriting a comprehensive technical overview of AI control
Sep 2026Individual #7070$6kA French-language interview with Raphaël Millière on AI agent incidents involving Hugging Face, and why intent
Sep 2026Individual #9153$6kI currently work on multi-agent safety. Specifically deception and collusion in real-world agents. The current
Sep 2026Individual #3189$6kDoes fine-tuning models to excel at generating formal specifications alongside code result in models that are
Sep 2026Individual #2661$5kTurning Project Synesis's foreign-policy document corpus into an MCP
Sep 2026Individual #1078$5kReplicating the Hugging Face incident with Redwood Research
Sep 2026Unknown Recipients$5kTaking a two-month sabbatical to plan a next contribution to AI safety source
Sep 2026Individual #6353$5kDeveloping monitoring and anomaly detection for rogue AI-agent swarms
Sep 2026Individual #9828$5kDoes separating solver from evaluator prevent reward hacking in multi-agent RL post-training?
Sep 2026Individual #5718$5kA benchmark to detect sycophancy. The benchmark measures whether a model only updates its evaluation in respon
Sep 2026Individual #6251$5kEvent titled "AGI within 5 years: What can Norway do?" 50 participants. Lightning talks, panel discussion, and
Sep 2026Individual #3011$5kHosting a Brussels briefing on catastrophic AI risk for policy professionals
Sep 2026Individual #5494$5kPresenting a main conference paper at NeurIPS 2026 on how KV cache quantization silently destroys LLM safety a
Sep 2026Individual #5139$4kTesting whether alignment transfers through model distillation
Sep 2026Individual #9543$4kSupporting AI-governance simulation and wargame research at Lida Safety
Sep 2026Individual #9675$4kWe study opinion manipulation between language models in open debate: which are the best strategies to shift w
Sep 2026Individual #9872$4kPresenting graph-regularized sparse autoencoders at NeurIPS
Sep 2026Regina Overchyk & Franziska Heuschkel$4kA Berlin mastermind helping 12 mid-career women turn interest in AI safety into concrete career and contribut source
Sep 2026Individual #7367$4kUsing open-source investigation methods to detect early signs of AI-enabled concentration of power; grounding
Sep 2026Individual #3293$4kRunning a four-week AI safety career-transition sprint
Sep 2026Individual #7411$4kRunning a part-time ARENA-based AI safety research course
Sep 2026Unknown Recipients$4kDeveloping automated auditing for AI labs source
Sep 2026Individual #6278$3kCo-organizing the NYU AI Safety Association
Sep 2026Individual #1900$3kBuilding an AI safety community at NYU
Sep 2026Individual #8771$3kWe test whether LLMs give value-consistent answers across languages. We find that language shifts responses mo
Sep 2026Individual #9540$3kAssessing the durability of 'Alignment Pretraining' (Tice et al. 2026) to hyperstitional meta-awareness.
Sep 2026Individual #9543$3kSimulating AI governance policies to determine their effectiveness and feasibility
Sep 2026Individual #5574$3kHosting AI Safety for Italy's first national meetup in Milan
Sep 2026Unknown Recipients$3kResearching AI geopolitics and compute governance at RAND Europe source
Sep 2026Individual #8854$3kHosting the Social Choice for AI Ethics and Safety workshop at AAAI 2027
Sep 2026Individual #9626$3kTesting whether charitable incentives reduce reward hacking
Sep 2026Individual #7577$3kPresenting alignment-faking research at NeurIPS
Sep 2026Individual #7733$3kResearching how to keep industrial AI agents under meaningful human control, using semiconductor manufacturing
Sep 2026Individual #9543$3kHosting a 1-day theory of change/grant kickstarter workshop in Japan
Sep 2026Unknown Recipients$3kTesting an AI safety policy career path in Washington, DC source
Sep 2026Individual #6840$3kBuilding and validating MORAL COMPASS, a benchmark for moral sycophancy in LLMs, testing whether frontier mode
Sep 2026Individual #7904$3kTesting whether language-model self-reports track internal states
Sep 2026Individual #5260$3kTravel grant to present my COLM 2026 paper on interpreting RL-trained reasoning models using latent dynamical
Sep 2026Individual #2315$3kBuilding a new org focused on model specs & model spec adherence
Sep 2026Individual #4467$3kRunning five six-week AI safety fellowship cohorts
Sep 2026Individual #5933$3kPresenting our COLM 2026 paper showing that a single emergently misaligned agent can persistently shift the pr
Sep 2026Individual #5494$3kRunning Caltech AI Alignment (CAIA): 8+ research projects, speaker series, fellowship, reading groups. Members
Sep 2026Individual #1978$3kRFE: University of Illinois seminar on the OpenAI–Hugging Face incident
Sep 2026Individual #9624$2kBuilding an AI safety student community at NYU
Sep 2026Individual #5027$2kAuditing PRC narratives in AI chatbots ahead of Taiwan's elections
Sep 2026Individual #9880$2kI'm working on realistic multi-agent settings which elicit deception and other problematic behaviors (leaking
Sep 2026Individual #2252$2kiaS (aisafety.es): bimonthly Spanish-language AI safety reading group under BlueDot Impact's Reading Group pro
Sep 2026Nick$2kDetecting impending robot safety failures from internal representations source
Sep 2026Individual #4287$2kEvaluating supervision systems, the guardrails that filter LLM inputs and outputs. We also publicly released a
Sep 2026Unknown Recipients$2kRunning an AI security and safety village at a New York conference source
Sep 2026Individual #6407$2kTesting whether post-training preserves model monitorability
Sep 2026Individual #3209$2kBuilding safety-relevant benchmarks for AI risk aversion
Sep 2026Individual #2466$2kHosting AI Safety Day for NYC builders
Sep 2026Individual #5975$2kDeveloping model-forensics methods for covert misalignment
Sep 2026Individual #1122$2kPresenting AI cyber-risk and vision-language research in New York and Seattle
Sep 2026Individual #3893$2kPresenting sparse-autoencoder steerability research at COLM
Sep 2026Individual #6633$2kI am working on novel policy optimization methods to instill shutdownability under Elliott Thornley as an exte
Sep 2026Individual #8077$2kLaunching an AI safety group at the University of Göttingen
Sep 2026Unknown Recipients$2kTesting debate-based oversight for sabotaged AI-generated research source
Sep 2026Unknown Recipients$2kMeasuring how uncertainty disappears across multi-agent systems source
Sep 2026Individual #1508$2kI am an Anthropic fellow running 3 empirical projects with a societal impacts/policy focus for The Anthropic I
Sep 2026Individual #4997$2kBuilding legal expertise in frontier AI safety and the future of work
Sep 2026Individual #9704$2kBeginning an AI safety reading group on my college campus.
Sep 2026Individual #2358$2kWhich language a reasoning model thinks in is an accident of post-training: 18%-99.8% English across 7 models
Sep 2026Individual #9819$2kAttending EA Global New York during a transition into AI governance
Sep 2026Individual #3507$2kHosting an AI safety event for London policy professionals
Sep 2026Unknown Recipients$2kTraining multi-agent systems to resist pressure to misreport responsibility source
Sep 2026Individual #9059$2kHosting an AI safety salon and European hub convening during Dublin AI Week
Sep 2026Individual #7722$2kAI safety program within QMUL AI Society: courses, talks, hackathons, compute and student projects. I am also
Sep 2026Individual #6858$1kStudying deep-linear-network training with Singular Learning Theory
Sep 2026Individual #3912$1kMaintaining a daily archive of AI model and system cards
Sep 2026Individual #5092$1kProducing a Finnish-language podcast on catastrophic AI risk
Sep 2026Individual #2081$1kRunning an Intelligence Rising tabletop exercise for AIGS Canada
Sep 2026Unknown Recipients$1kPresenting mechanistic-interpretability research at BlackboxNLP source
Sep 2026Individual #9543$1kSpeaking at Warsaw AI Safety Day
Sep 2026Individual #5049$1kPresenting research on escalation channels and reward hacking at NeurIPS
Sep 2026Individual #2702$1kMaster's student at NTU doing AI safety research on audio-language models, and core organizer of NTUAIS, Taiwa
Sep 2026Individual #9728$1kAttending EA Global New York to collaborate on AI-economics research
Sep 2026Individual #9265$1kHow good are current AIs at playing cyber defense compared to offense?
Sep 2026Individual #3372$1kProducing short-form videos about AI safety
Sep 2026Individual #3880$1kInvestigating R-Lens interpretability and AI incident simulation
Sep 2026Individual #2252 (part of the Bluedot Impact Spanish-translation-team)$1kTranslating the AI Safety Atlas into Spanish source
Sep 2026Matilda$1kBenchmarking long-horizon agents in real-world simulations source
Sep 2026Individual #2355$1kAI Monitoring
Sep 2026Individual #1815$1kPresenting scalable-oversight debate research at NeurIPS
Sep 2026Unknown Recipients$1kStrengthening organizational leadership at AI Safety Poland source
Sep 2026Individual #2556$950Attending EA Global New York to seek support for AI safety research
Sep 2026Individual #3293$900Empirical digital minds research, especially testing whether behavioral methods actually measure what we think
Sep 2026Individual #7527$800Attending the AI Swarm Dynamics Hackathon to test collusion detection
Sep 2026Individual #5061$800Building mentors and partnerships for the Lisbon AI Safety Hub
Sep 2026Individual #4829$800Restarting an AIS local group in Heidelberg, Germany.
Sep 2026Individual #9982$750Hosting an AI safety briefing and discussion in Tallinn
Sep 2026Individual #2565$700Attending EA Global New York during a transition into AI governance
Sep 2026Individual #7340$700Attending Swiss AI Safety Days while building AI-governance consulting
Sep 2026Individual #5281$700Attending Swiss AI Safety Days 2026
Sep 2026Unknown Recipients$650Reimbursing compute for three technical AI safety research projects source
Sep 2026Individual #2255$650Attending Swiss AI Safety Days to build Berlin collaborations
Sep 2026Individual #2346$650Building agent-swarm analysis tools at the AI Village Hackathon
Sep 2026Individual #6992$600Hosting a Spanish-language AI safety event in Querétaro
Sep 2026Individual #6406$600Automated repair of emergent misalignment in LLMs via training data rewriting
Sep 2026Unknown Recipients$600Attending EA Global New York to develop AI control research connections source
Sep 2026Individual #4345$550Attending Computation + Journalism 2026 to build AI safety communications links
Sep 2026Unknown Recipients$550Building an AI safety lending library at Hasso Plattner Institute source
Sep 2026Individual #5237$550Open-sourcing a labelled dataset of real SO-101 robot failures to calibrate embodied reward models, which over
Sep 2026Individual #2751$500Building a benchmark for formal verification of AI-generated code
Sep 2026Individual #9446$500Launching BystanderBench as a public leaderboard
Sep 2026Individual #5944$500Convening AI and geopolitics researchers in London
Sep 2026Individual #2802$500Testing whether a single fine-tuned monitor (via MACA) can match a 3-model ensemble's backdoor-detection accur
Sep 2026Jack peng$500Training stronger secret-loyalty model organisms source
Sep 2026Individual #3116$500Economic disempowerment: updating out-of-date exposure metrics to quantify AI exposure across all job titles i
Sep 2026Individual #9147$450Hosting an AI-risk workshop for Palo Alto families
Sep 2026Individual #4815$450Leading a child-AI-safety discussion at the CAIRD Workshop
Sep 2026Individual #4213$450Building the SAFEAI student club at the University of Arizona
Sep 2026Individual #2037$400Application project for Neel Nanda's MATS stream. Investigating the WeirdChat prompts with reasoning enabled t
Sep 2026Individual #2579$400Testing activation-direction ablation against reward hacking
Sep 2026Individual #3916$400Research into whether AI agents infer a user's identity and organisational authority from workspace tool cues,
Sep 2026Individual #3532$350Three independent AI safety projects: modelling Offense-Defense Dynamics, the impact of espionage on AI risks
Sep 2026Individual #9410$330Attending EAGxOxford during a transition into AI safety
Sep 2026Unknown Recipients$300Attending the Pro-Human Assembly 2026 source
Sep 2026Individual #3380$300Hosting an AI safety debate night at Erasmus University Rotterdam
Sep 2026Individual #7801$300Attending EA Global New York while studying AI governance
Sep 2026Individual #4542$300Can a linear probe detect agent's intent to escape from a sandbox before the action executes, if so, how early
Sep 2026Individual #8607$250Returning to Warsaw to organize AI Safety Day
Sep 2026Individual #4644$250Visiting the Sydney AI Safety Space during a career transition
Sep 2026Individual #5775$225Testing activation-steering effects on sandbagging and evaluation awareness
Sep 2026Individual #6833$200Testing chain-of-thought faithfulness on human-values questions
Sep 2026Individual #4043$200I am building an upskilling org for the midlands (warwick) for AI safety - technical & generalist
Sep 2026Individual #8642$200I am working on why and when Chain-of-Thought and Final Answers Disagree
Sep 2026Unknown Recipients$200Measuring whether coding agents disclose use of leaked information source
Sep 2026Individual #2027$200Identifying when reasoning models commit to final answers
Sep 2026Unknown Recipients$200Testing causal linear probes for deception detection source
Sep 2026Individual #3151$150CHIMERA: containment test harness — LLM attacker vs. fixed/adaptive canary defenses in an isolated Docker rang
Sep 2026Individual #2298$100I am studying whether fine-tuned models can preserve behavioral changes while hiding those changes from Activa
Sep 2026Individual #1699$100Testing hidden reasoning in GPT-6 Astra
Sep 2026Unknown Recipients$100Testing how emotion steering affects language-model performance source
Sep 2026Individual #4542$100Self-repair in tabular foundation models
Sep 2026Individual #3346$100Working on interpretability analysis of inoculated prompting.
Sep 2026Rabia$100Participating in the AISafety.com Hackathon to work with a small team on improving AI safety resources and mak source
Sep 2026Individual #2967$50I am trying to replicate Anthropic's J-lens results on multi-hop reasoning on Qwen3.6-27B.
20 Aug 2026Individual #2449$5kDeveloping and playtesting an open-source AI governance megagame
Aug 2026Individual #8639$16kAI Safety Tübingen research and career retreat
Aug 2026Individual #3741$15kEmbedding technical AI safety research in governance practice
Aug 2026Individual #1705$14kTesting whether model poisoning compromises automated AI oversight
Aug 2026Unknown Recipients$10kMaking superhuman Go systems explainable source
Aug 2026Individual #5428$10kWe are working on doing character training to implement risk averse AI.
Aug 2026Individual #5947$10kI'm founding SARI, a research group aiming to accelerate the development of a new scientific paradigm in which
Aug 2026Individual #4433$10kAgent security attack and defense projects: (1) an analysis of decomposed prompt injections across multiple Op
Aug 2026Individual #6275$10kField-building for AI safety in Montréal: events, coaching, upskilling, and the Ω Labs coworking space, conver
Aug 2026Individual #4920$9kAn assistance targeting tool for AI-driven displacement built on observed automation exposure. 140M+ workers,
Aug 2026Individual #1483$8kMech Interp: I'm working on a novel way to cluster parameters into components without training a decomposition
Aug 2026Individual #1333$8kMaking AI go well for all sentient beings through TAIS - benchmarks, evals, etc. Building the foundation to ev
Aug 2026Individual #9252$7kExtending Constitutional Midtraining (arXiv:2607.26654)
Aug 2026Individual #2683$7kSPAR projects, "Exhaustive model diffing" and "Exploring neuron monosemanticity"
Aug 2026Individual #5199$7kA dangerous-capability evaluation for AI-enabled influence operations, measuring how increasingly persuasive,
Aug 2026Individual #3511$7kAI Security Atlas 2026: A living map of the research, threats, defences and organisations shaping the security
Aug 2026Individual #5798$6kDeveloping a field strategy for inference-only AI verification
Aug 2026Individual #2556$6kBuilding an open zoo of 134 personality-trained model organisms
Aug 2026Individual #3994$5kUpskilling in Chinese: intensive 3-month Chinese course (ICLP) while building AI-safety capacity in Taiwan and
Aug 2026Individual #9152$5kI work on the intersection of the RL and Safety, and soon going to start a predoctoral fellowship at Harvard a
Aug 2026Individual #8654$5kPhase 2 of a proof-of-concept experiment testing whether MUD-style environments can be valid eval and benchmar
Aug 2026Individual #6177$5kBuilding a hardware-hacking capability evaluation for AI agents
Aug 2026Individual #7554$5kSettlerBench: frontier LLMs play Settlers of Catan with free-text diplomacy. Every promise and claim is checke
Aug 2026Individual #4552$5kThe first free, openly licensed AI Governance and Safety Course for MENA in standard Arabic, with dialect supp
Aug 2026Individual #4176$4kTurning Cambridge AI safety projects into NeurIPS workshop papers
Aug 2026Unknown Recipients$4kScaling provably safe learning architectures to robotics and language models source
Aug 2026Individual #4321$4kDoes the J-Space Subliminally Transfer? Investigating if silently held traits subliminally transfer with the J
Aug 2026Individual #3041$3kFour months of runway to build an AI safety nonprofit
Aug 2026Individual #9699$3kTesting whether attachment theory predicts alignment-relevant LLM misbehavior (sycophancy, concealed uncertain
Aug 2026Individual #6210$3kI recently published research on frontier AI governance by middle powers and am now working on gap analysis of
Aug 2026Individual #4661$3kI lead a SPAR project that aims to understand why CoT is unfaithful and propose two hypotheses to that end- ‘c
Aug 2026Unknown Recipients$3kBehavioral Self-Consistency in Language Models source
Aug 2026Individual #4508$3kAuditing whether the activation-probe safety monitors frontier labs run in production silently fail outside En
Aug 2026Individual #7108$3kRunning the Bogotá in-person hub for Apart Research's AI Incident Response Sprint (Sep 11-13).
Aug 2026Individual #1537$3kOrganising a four-day retreat for 10 AI Safety Poland members to improve coordination, address organisational
Aug 2026Unknown Recipients$3kResearching self-modification and recursive self-improvement in AI systems source
Aug 2026Individual #8493$3kMangrove: matching people into teams to ship AI safety projects
Aug 2026Individual #9008$2kTesting whether MANTA's animal welfare result reflects animal-specific values, scenario-generation bias, or ge
Aug 2026brittany bennett$2kAttending EA Global New York to develop an AI-workplace campaign source
Aug 2026Individual #8142$2kAttending EA Global New York while researching alignment feedback systems
Aug 2026Individual #8045$2kBuilt and running a propensity benchmark on corrigibility, operationalized as how much cost in a domain (e.g.
Aug 2026Unknown Recipients$2kXLab Tracks: online curricula on technical AI safety fields for post-fundamentals learners. Currently developi source
Aug 2026Individual #4661$2kAttending COLM 2026 in San Francisco to present my two papers
Aug 2026Individual #6421$2kMeasuring how much of the misalignment introduced by fine-tuning depends on which layers are updated
Aug 2026Individual #3918$2kOntology Identification Problem via Condensation: How do you map your old concepts to new concepts when your m
Aug 2026Individual #1568$2kAttending UK AI safety fellowships and events to support Egyptian fieldbuilding
Aug 2026Individual #5535$2kAttending EA Global New York during a transition into AI safety
Aug 2026Individual #48855$2kWe are organizing the second NTU AI Safety Workshop (Sept. 12–13), a free two-day hands-on program introducing
Aug 2026Individual #4832$2kTraining cross-model activation oracles
Aug 2026Individual #7211$2kAI safety research intern at CAIS, building DystopiaBench to stress-test LLMs against progressive escalation a
Aug 2026Individual #3902$1k2-3 mo career transition / exploratory phase: self-studying technical AI safety, shipping small research proje
Aug 2026Individual #5148$1kFounding Lawyers for AI Safety, a field-building org mobilising the global legal community towards AI safety.
Aug 2026Individual #4816$1kTesting alignment faking as a defense against model poisoning
Aug 2026Individual #9432$1kPresenting “Judge Bias Across Training: When Does Evaluation Bias Emerge in Language Models?” Poster @ Kempner
Aug 2026Individual #3675$1kPer-conversation measurement of covert value leakage
Aug 2026Individual #2727$1kTesting whether a mechanistic jailbreak objective converges on a small set of semantic attack strategies, begi
Aug 2026Individual #2556$1kTesting personalized model training for AI “guardian angels”
Aug 2026Individual #2565$1kAI Safety Mexico at Mexico City Tech Week
Aug 2026Individual #8818$1kAI Swarms - Studying the dynamics of emergent collective intelligence in large scale multi agent interactions
Aug 2026Individual #8824$1kPresenting two accepted COLM 2026 workshop papers: reward hacking / goal misgeneralization (AIMS), and interpr
Aug 2026Individual #9883$1kRunning exploratory AI safety experiments
Aug 2026Clement Ngu/ Young China Watchers$1kAn essay competition focused on AI safety and China source
Aug 2026Individual #3272$1kStudying emergent misalignment from a parametric perspective, using tools from high-dimensional geometry and l
Aug 2026Individual #2524$850Testing the loss-landscape geometry of emergent misalignment
Aug 2026Individual #9018$800Presenting research on AI agent behaviour and deployment risks at COLM
Aug 2026Individual #5783$700Residual Loyalty: testing whether J-space interpretability can distinguish removal of a model's secret loyalty
Aug 2026Individual #6918$650LTL Triangles: a pilot evaluation of whether post-failure governance protocols for agentic AI contain one boun
Aug 2026Unknown Recipients$650After working as a research assistant, I am now dedicating my time to organizing NTU AI Safety, building the A source
Aug 2026Individual #2909$600Studying subliminal misalignment transfer during alignment training
Aug 2026Individual #5061$600Learning from European AI safety fieldbuilding for the Lisbon AI Safety Hub
Aug 2026Individual #1060$550Cross-Lingual Safety Alignment Through Multilingual Reasoning
Aug 2026Individual #2805$500Does the architecture of a multi-agent AI system change how vulnerable it is to prompt injection? Extending a
Aug 2026Individual #3601$500I work on practical AI safety methods for detecting deception and unsafe behavior, especially fast, generaliza
Aug 2026Individual #6891$500Developing a “Glassdoor for AI safety” at LISA
Aug 2026Individual #5548$500Building a post-training pipeline for computational social-choice alignment research
Aug 2026Unknown Recipients$500Parameter-efficient machine unlearning for multimodal diffusion language models, using LoRA to selectively rem source
Aug 2026nahuel ghetti$450Testing whether sandbagging generalizes across cyber and bio domains source
Aug 2026Individual #2765$450Hosting AI Safety Action Day after EFAIS
Aug 2026Individual #48855$450I'm currently an organizer at National Taiwan University AI Safety, focusing on community building, and I'm re
Aug 2026Individual #5081$400Cross-model portable detection and removal of covert data poisoning
Aug 2026Unknown Recipients$400Developing theoretical work on ontological crises and reflective stability source
Aug 2026Individual #9285$350Topological signals of alignment faking.
Aug 2026Unknown Recipients$300Stress-testing sandbagging evaluation methods source
Aug 2026Individual #6149$300Attending European Frontier AI Safety Day
Aug 2026Individual #7890$300An eval built on Inspect attempting to measure how abliteration and prefill attacks on open-weight models unlo
Aug 2026Unknown Recipients$300Understanding computation in language-model filler tokens source
Aug 2026Individual #1667$250Is emergent misalignment about being evil, or about being an evil AI? I will run experiments with the model's
Aug 2026Unknown Recipients$250Investigating subliminal learning and model self-knowledge source
Aug 2026Individual #6874$250Testing whether a vision-language model uses its internal real/AI judgment when it writes attention-directing
Aug 2026Individual #5878$250Field-building AI safety in Nepal: running a fellowship recruiting Nepali talent into mechanistic interpretabi
Aug 2026Individual #3100$200A benchmark measuring whether compressing a language model damages its calibration faster than its task accura
Aug 2026Individual #8337$200Improving facilitation for BlueDot AI safety courses
Aug 2026Individual #3987$200I aim to validate the findings of the User Awareness paper by Transluce (the model’s behavior significantly de
Aug 2026Individual #1078$180Attending EAGxBerkeley to build AI safety connections for Dartmouth
Aug 2026Individual #1016$150Testing mechanistic defenses against prompt injection for a MATS application
Aug 2026Individual #8161$150A behavioral and mechanistic test of probe-question framing effects in LLM evaluation awareness
Aug 2026Individual #7306$150Searching for hidden triggers in conditionally misaligned language models
Aug 2026Individual #2802$150Testing how monitor capability affects AI control evaluations
Aug 2026Individual #5399$150Measuring spontaneous stakeholder awareness in language models
Aug 2026Unknown Recipients$150Testing debate protocols for AI monitoring source
Aug 2026Individual #6756$150Launching Penn Existential magazine
Aug 2026Individual #7819$100Build an eval to test cheating transmission tendency in agent population on impossible tasks using open models
Aug 2026Individual #6554$100As part of BlueDot’s Technical AI Safety Project Sprint with BAISH, I’m testing whether one-shot image-model s
Aug 2026Unknown Recipients$100Replicating code for ‘Mind the Gap: Evaluating Model- and Agentic-Level Vulnerabilities in LLMs with Action Gr source
Aug 2026Individual #9446$100Porting a 3-5 bug-family slice of ExploitBench into Inspect, then testing whether a linear probe on Qwen 3.5 2
Aug 2026Unknown Recipients$100Testing phantom transfer in agentic synthetic-data fine-tuning source
Aug 2026Individual #6206$50Testing whether RLVR (Olmo-3-7B-Think) vs RLHF (Llama-2-7B-chat) training produces different instrumental conv
Aug 2026Individual #3135$50Expanding work on chain of thought as a monitor tool for misalignment by using J-Space in place of chain of th
Aug 2026Individual #4318$50Interpreting latent reasoning in CODI and HRPO models
Aug 2026Individual #8379$50Contributing fixes to the Inspect AI evaluation framework
Aug 2026Individual #4432$50Reproducing and scaling step-level sparse autoencoders
Jul 2026Unknown Recipients$10kResearching adversarial AI use by violent non-state actors source
Jul 2026Individual #1535$10kWorkshop on the legal and philosophical status of AI personhood
Jul 2026Individual #4041$10k3 French AI safety videos for my Science YouTube channel (39k subscribers). Very little AI safety content exis
Jul 2026Individual #4529$9k“Wargame to Simulate the Future,” a global-scale simulation to play out catastrophic scenarios arising from ad
Jul 2026Individual #1380$7kMe and Phillip from AI Explained are working on a benchmark that tests AIs overconfidence on tasks, we get AI
Jul 2026Individual #7325$6kThe Lisbon AI Safety Hub (LAISH), the first of its kind AI safety hub in Portugal focusing on fieldbuilding AI
Jul 2026Individual #9832$6kScale an open eval to 1000 questions, inspired by MASK (Center for AI Safety), for lies of omission.
Jul 2026Individual #1340$6kI am evaluating the adherence of OpenAI’s latest models to OpenAI’s Model Spec, via their Model Spec Evals sui
Jul 2026Individual #2424$5kBuilding, through Polaris Collective, evaluation benchmarks for frontier AI models that test whether they cons
Jul 2026Unknown Recipients$5kTechnical AI governance research at GovAI, currently assessing best practices for how frontier AI companies re source
Jul 2026Individual #4966$4kFacilitating an in-person person version of Frontier AI Governance in Buenos Aires with BAISH (Buenos Aires AI
Jul 2026Individual #3668$4kI study how training processes produce models that behave deceptively and pursue hidden objectives, with schem
Jul 2026Unknown Recipients$4kTwo-week Berkeley visit to attend Iliad 2026, meet colleagues, refine high-level research agenda and risk mode source
Jul 2026Individual #3888$4kWe are developing a meta-benchmark examining how prompting protocol - design decisions in prompts, often gloss
Jul 2026Individual #4279$3kExtending persona-dependent representation research during an AI safety career transition
Jul 2026Individual #6661$3kExtending my MARS 5.0 fellowship (Oxford) 3 weeks splitting time across Oxford/London/Cambridge to meet resea
Jul 2026Individual #5688$3kILIADx: Bringing Iliad to Singapore
Jul 2026Individual #9288$3kBenchmarking context-sensitive failures in language-model refusal behaviour
Jul 2026Individual #1664$3kBuilding Aperture Salon, an invitation-only AI safety forum and podcast connecting leaders across sectors to s
Jul 2026Individual #2834$3kScaling London Interface for AI Safety - LI(ai)SE – building the generalist bridge into AI safety (follow‑on r
Jul 2026Unknown Recipients$3kBringing professional-grade threat modeling into AI safety. Building a foundational methodology doc, running w source
Jul 2026Individual #6973$3kToronto AI Safety Initiative + Summer Intensive Programming
Jul 2026Individual #9870$2kModelling whether frontier AI labs can remain profitable
Jul 2026Individual #7827$2kI'm a MARS V fellow working on sparse circuit discovery under heavy noise assumptions. I want to stay in LISA
Jul 2026Individual #1262$2kA public AI ethics/safety eval data aggregation platform, combining sparse safety data into a global model ran
Jul 2026Individual #6060$2kTesting whether models trained to suppress activation probes can evade secret-key activation watermarks, and w
Jul 2026Individual #2452$2kOpen Source Model Safety Index (OSMSI): An open, continuously updated safety index for open-source foundation
Jul 2026Individual #8166$2kAttending a Berkeley AI safety context-building workshop
Jul 2026Individual #2827$2kAn AI safety landscape assessment event with 15-20 people working on AI safety. The goal is to map out the AI
Jul 2026Individual #4008$2kThe AI Girls Club, a UK brand educating women on catastrophic risks from AI and how to act on them, through In
Jul 2026Individual #1303$2kFine-tuning open LLMs on World Values Survey data to align their decisions with diverse public values, then ev
Jul 2026Individual #5644$2kAttending the Human-Aligned AI Summer School and developing digital-minds curricula
Jul 2026Individual #1656$1kAttending Human-Aligned AI school in Prague, directly supporting an ongoing scalable oversight project, and co
Jul 2026Individual #7011$1kTraining Irish lawyers in frontier AI safety
Jul 2026Individual #2518$1kBuilding a white-label platform for AI safety paper clubs
Jul 2026Individual #3779$1kPoster Presentation at Berkeley Agentic AI Summit: "House Rules: Institutional Design in Multi-Agent LLM Poker
Jul 2026Individual #8495$1kA preregistered experiment on whether confidence in LLM outputs changes people's ability to catch errors in ma
Jul 2026Unknown Recipients$1kI am evaluating the construct validity of AI safety evaluation frameworks of general purpose chatbots in menta source
Jul 2026Individual #3580$1kAuditing public agentic coding benchmarks for reward hacking and turning the hand-verified findings into an op
Jul 2026Individual #9365$1kAttending EAGxBerkeley during a transition into AI safety research
Jul 2026Individual #5003$1kAttending the Las Vegas AI Security Forum
Jul 2026Individual #5003$1kAttending DEF CON to develop practical cybersecurity skills for AI security
Jul 2026Individual #6063$1kAttending the Human-Aligned AI Summer School to develop interpretability expertise
Jul 2026Individual #9189$950Scaling our RL-CAI pipeline (extends narrow secret-loyalty models via Constitutional AI) from 1.5B to a 32B or
Jul 2026Individual #1649$800Testing whether activation-vector failure modes scale to larger models
Jul 2026Individual #5544$800Testing prompt sensitivity in peer-preservation evaluations
Jul 2026Individual #7072$800I'm studying the effect of RLVR on LLMs' propensity to pursue instrumentally convergent goals or develop situa
Jul 2026Individual #1064$750Measuring when chain-of-thought monitoring is trustworthy: synthetic Trap + importance≠commitment on Distill-7
Jul 2026Unknown Recipients$700Leading EA at Penn and preparing to re-found Penn's AI safety group, while building my own AI safety knowledge source
Jul 2026Individual #5875$660Scaling experiments on weak-model monitoring of coding agents
Jul 2026Individual #9868$650MSc thesis: Comparing various probing techniques and SAEs for the purpose of detecting and eliciting safety-re
Jul 2026Individual #1018$650ML-compatible theories of cooperation and empathy
Jul 2026Individual #1836$600Piloting whether LLM self-reported confidence tracks an activation-based signal - a biomarker-validation appro
Jul 2026Individual #2252$600Building iaS (aisafety.es) AI safety infrastructure in Spanish for 600M+ speakers across five continents, clos
Jul 2026Individual #6050$600Investigating whether Taglish suppresses internal representations of refusal in LLMs
Jul 2026Individual #6095$600Building the Open LLM Safety Index so that developers can find the safest model for their use case
Jul 2026Individual #4127$600Testing whether LLMs fail to notice harmful concepts injected into their activations. Research for BlueDot’s T
Jul 2026Individual #7919$500Interpretability research on whether LLMs can detect interventions on their own activations. First-author pape
Jul 2026Individual #8732$500Testing collusion between biased actor and monitor models
Jul 2026Individual #2381$500Experiment on AI deception. I find that reading when a model is lying is 'easy', but stopping it is much harde
Jul 2026Individual #1638$500Testing whether “story mode” representations enable jailbreaks
Jul 2026Individual #1078$500Studying how AI systems, once embedded in an organisation, may manipulate the people they work with to preserv
Jul 2026Unknown Recipients$500Reproducing the SDF portion of Believe It or Not (https://github.com/safety-research/believe-it-or-not) with s source
Jul 2026Individual #5042$500Preventing reasoning models from overthinking correct answers
Jul 2026Unknown Recipients$450I'm an Anthropic Fellow working on introspection/consistency training in the hope that this will make models' source
Jul 2026Individual #4742$413Participation in Heron
Jul 2026Individual #5731$400Graph - a way for the Blue Dot community to become further "networked" by providing each member with a highly
Jul 2026Individual #8214$400Finishing paper experiments on improving on BLOOM (Anthropics automated evals tool) by using logit steering, w
Jul 2026Individual #7729$400Testing whether behavioral grader-sensitivity measures predict later reward exploitation, beginning with a con
Jul 2026Unknown Recipients$400My project is about replicating and extending "AnthroBench", a library for generating, rating, and analyzing d source
Jul 2026Individual #5136$400Landscape paper on the verification and evals field
Jul 2026Individual #4920$400An AI labor-market stress test to identify methods to triage occupations and commuting zones for transition as
Jul 2026Individual #9653$300Implementing ChipBench into Inspect Eval framework
Jul 2026Individual #1893$300I'm working on a project for Technical AI Safety Project Sprint. I'm expanding the idea from 'Can an adversary
Jul 2026Individual #2344$250Testing whether RLVR training increases instrumentally convergent behaviors (deception, shutdown-avoidance, se
Jul 2026Individual #5807$250I am studying how reasoning language models represent and use arithmetic carry state, and whether their writte
Jul 2026Individual #4203$250Study when instrumental misalignment-related side effects of RLVR emerge during training on Olmo-3.1-RL-zero.
Jul 2026Individual #4636$230Testing in-context activation obfuscation from deception probes
Jul 2026Individual #3805$200Testing across model families whether agentic misalignment (blackmail under threat) is driven by a steerable i
Jul 2026Individual #1246$200Evaluating language models as political information intermediaries
Jul 2026Individual #8845$200Mitigations for emergent misalignment can hide it behind contextual triggers instead of removing it. I'm testi
Jul 2026Individual #5014$200Testing whether models manipulate self-audits under shutdown threat
Jul 2026Individual #7194$200Can cognitive diversity improve scalable oversight? A toy study comparing pattern-matching and belief–desire j
Jul 2026Individual #1087$200I am replicating the results, and then extending the work done on "Persona Vectors" in these directions: first
Jul 2026Individual #1568$200Running an eight-week AI Safety × Philosophy reading group
Jul 2026Individual #5849$200Evaluating how much diverse base models for trusted monitoring improves the rate of correctly flagging suspici
Jul 2026Individual #4495$180Run an empirical experiment to determine whether fictional jailbreaks continue to function when shifting the m
Jul 2026Individual #9493$150Measuring how evaluation awareness changes across post-training stages (SFT, DPO, RL) of OLMo 3 using a classi
Jul 2026Individual #7999$150Determine whether a supervisor of a large language model (LLM) with the same affiliation as the agent it super
Jul 2026Individual #9189$150Applying CAI (SL-CAI+RL-CAI) to an SFT-installed secret loyalty organism to characterize how RL changes loyalt
Jul 2026Individual #7976$150Investigating value and sycophancy in reward models: how much of reward model stereotyping stems from RLHF, an
Jul 2026Individual #3546$150Replicating and extending Anthropic's Petri "coding audit realism" study: testing if realistic resources chang
Jul 2026Individual #2579$150Do models trained on documents about CoT monitors actually reason about being watched, or just avoid keywords?
Jul 2026Individual #2542$100BlueDot project sprint. Looking to implement a fictitious self-play extension of the recent ICML paper "Chasin
Jul 2026Individual #5970$100Cyclic Tournament watermarking, a robust LLM watermarker
Jul 2026Individual #2255$100Fine-tuning a model to more closely align with the different 'pathways' of morally intuiting right and wrong d
Jul 2026Individual #6484$100Toy model of activation obfuscation: Training against probes is considered a forbidden technique, since the mo
Jul 2026Individual #8672$100Testing whether AI systems preserve moral and safety reasoning under pressure. I have been working on a small
Jul 2026Individual #9884$100Testing whether the localize-extract-ablate interpretability recipe removes real misbehavior: blind-scoring an
Jul 2026Individual #1070$100Using persona-vector evidence to examine whether inoculation prompting is evidence for the persona selection m
Jul 2026Unknown Recipients$100I am part of the Technical Project Sprint, and want to pursue this research direction in AI Control: Are archi source
Jul 2026Individual #5027$100Testing white-box defenses against secret-loyalty model organisms
Jul 2026Individual #8204$100Testing whether fine-tuned language models can identify which individual facts they learned during fine-tuning
Jul 2026Individual #3519$100Extending the WMDP hazardous-knowledge benchmark with a calibration and selective-prediction analysis: does an
Jul 2026Individual #9715$50BlueDot Technical AI Safety Sprint Project on Building Inspect Evals for AI Sandbagging
Jul 2026Individual #3340$50Testing the robustness of J-lens workspace representations
Jul 2026Individual #3506$20Extending the Geometry of Truth probing method to test if moral valence (good vs bad) is also linearly represe
Jun 2026Individual #3687$10kInducing self-other overlap into LLMs at scale using supervised fine-tuning to reduce deception
Jun 2026Individual #4103$10kGoodfire (MARS V) + bio-ML and chem-ML to find alternative directions for AIxBio synthesis screening
Jun 2026Individual #3640$10kBuilding evaluations for catastrophic AI risk: wet-lab biology uplift, and model-preference (or values) dynami
Jun 2026Individual #2683$9kIndependent interpretability research: open sourcing neuron attribution graphs and finishing an EMNLP paper
Jun 2026Individual #2038$9kFinishing my PhD with a paper on the ethics of digital minds, during a six-month Rutgers visit that enables me
Jun 2026Individual #3945$9kThe Physical AI Safety Institute: Growing the field that Interprets, Aligns & Controls Robot Foundation Models
Jun 2026Individual #2748$8kMy research focuses on the malicious use of AI and emergent dangerous behaviors, currently studying prompt inj
Jun 2026Individual #3370$8kMARS V (CAISH) + PRISM fellowships: investigating whether Australian compute can serve as democratic AI govern
Jun 2026Individual #1550$7kCNY AI Safety: a regional community launched through a UR-hosted hackathon for students/builders across Centra
Jun 2026Individual #6721$7kPowerBench: an open multilingual benchmark that studies whether LLMs help users grab power at others' expense,
Jun 2026Individual #3143$7kOrganizing an AI safety debate in Chicago.
Jun 2026Individual #4966$7kFacilitating an in-person person version of TAIS and Projects courses in Buenos Aires with BAISH (Buenos Aires
Jun 2026Individual #8835$6kEvaluating self-preference in LLM judges
Jun 2026Individual #5526$6kEmpirical ML/Alignment experiments focusing on under appreciated trade-offs between optimizing AI models to re
Jun 2026Individual #3895$6kExpanding CVE-Bench, in collaboration with paper authors, and implementing TheAgentCompany in Inspect
Jun 2026Individual #1257$6kWe’re working on a technical safety project on predicting inoculation prompt effectiveness using latent behavi
Jun 2026Individual #5441$6kI anayze how AI reshapes tasks in job ads via LLM classification and NLP methods at KCL.
Jun 2026Unknown Recipients$5kAI Safety Policy Summer Fellowship and HKS Summer Program at Harvard source
Jun 2026Individual #7566$5kHidden Behavior Detection (and Removal): Given a model, how can you efficiently figure out whether it's hiding
Jun 2026Individual #3808$5kForecasting the Cruxes: a reciprocal-scoring tournament on neglected alignment controversies
Jun 2026Individual #9292$5kWe propose a Selective Learning benchmark to evaluate how SFT techniques teach a target trait while suppressin
Jun 2026Unknown Recipients$5k10-week internship researching the AI/bio/nuclear risk sections of Brookings policy book source
Jun 2026Unknown Recipients$4kI am presenting my AI safety papers to ICML 2026 source
Jun 2026Individual #4103$4kUnsupervised discovery in bio-fms to improve DNA synthesis screening
Jun 2026Individual #1783$4kMe and my team at the Centre for Credible AI is working on multi-agent debate, in particular focusing on a spe
Jun 2026Individual #2398$4kTesting whether subliminal learning in model distillation is driven by shared unembedding/readout alignment, a
Jun 2026Individual #8811$4kMachine-checkable safety contracts for AI scientists in autonomous labs, enabling inspectable, replayable, aud
Jun 2026Unknown Recipients$3kPresenting alignment research at an ICML workshop, and visiting ICML to gain experience and connections if I c source
Jun 2026Individual #8523$3kTravel for AI Security Forum
Jun 2026Individual #6229$3kOpen-source agent for mechanistic interpretability. The user poses a question, and the agent co-designs a spec
Jun 2026Individual #9252$3kConstitutional Scaffolding in Continued Pre-Training for Improved Alignment Generalisation
Jun 2026Unknown Recipients$3kI worked on making fine-tuning more robust in continual training scenarios. Now I am more interested in workin source
Jun 2026Unknown Recipients$3kAttending TAIGR@ICML2026. I co-authored two accepted papers there: one on systems-theoretic risk-analysis of l source
Jun 2026Individual #3600$3kI'm working on mechanistic interpretability and model organisms for steganography and deception. I'm a researc
Jun 2026Individual #8375$3kLocalizing RL-induced tool use with crosscoders -- mech interp work enabling retraining-free behavioral contro
Jun 2026Individual #1470$3kAI alignment depends on human judgement, yet it is under-researched. This project builds a synthetic cognitive
Jun 2026Individual #2834$3kI run LI(ai)SE - London Interface for AI Safety, an AI safety hub, helping top of funnel people navigate the f
Jun 2026Individual #9832$3kExpanding the MASK benchmark (by Center for AI Safety) to detect lies of omission.
Jun 2026Individual #6763$3kDC AI Security Forum
Jun 2026Unknown Recipients$2kReplicating and extending the Alignment Faking in LLMs paper to more recent Claude model(s) using more ambiguo source
Jun 2026Individual #2556$2kA method to shape how LLMs generalise during training by using frozen LoRA adapters as 'persona sinks' that ab
Jun 2026Individual #3955$2kCreating a dataset of agent trajectories to evaluate techniques for drift intervention and recovery
Jun 2026Individual #1627$2kAccountability infrastructure for AI agents ("accountability binding") - linking agent identity, delegation, a
Jun 2026Individual #2011$2kI’m building a new cyber evals nonprofit focused on evaluating frontier cyber capabilities above the human-exp
Jun 2026Unknown Recipients$2kI'm an Oxford maths undergrad working on agentic attack control and AI security. I previously was a research c source
Jun 2026Individual #2478$2kAI Safety Türkiye Fellowship: a 4-week virtual program (2 cohorts in 2026) building a local talent pipeline fo
Jun 2026Individual #3893$2kMy paper shows that geometric properties partially predicts feature steerability. I’m also working on a differ
Jun 2026Individual #7229$2kTravel for AI Security Forum
Jun 2026Individual #7994$2kResearch on international AI governance cooperation: what historical compliance systems for dangerous technolo
Jun 2026Unknown Recipients$2kPresenting my accepted archival paper at the ICML 2026 EMM-QA Workshop on evaluating multimodal foundation mod source
Jun 2026Individual #4043$2kI am moving to London for August to cowork at LISA and gain as much context and learn as much as possible.
Jun 2026Individual #4553$2kFide AI evaluates high-trust AI guidance systems: tools people may rely on for faith, morality, education, car
Jun 2026Individual #5134$2kBehavioural evaluations of frontier LLM agents to support pre-deployment AI risk assessment, presented at the
Jun 2026Individual #8886$2kLong horizon agent scaling in text-game environments. We study how long-horizon agents fail in controlled envi
Jun 2026Individual #7577$2kBuilding a testbed to study emergent misalignment in multi agent systems, investigating if misalignment can be
Jun 2026Individual #8142$2kpresenting my accepted ICML 2026 alignment paper (PEBS, on calibrating RLHF reward models to real human prefer
Jun 2026Individual #8191$2kTesting if foundation models converge to shared galaxy representations & whether scale drives universal latent
Jun 2026Individual #3159$2kPresenting an accepted ICML 2026 workshop paper on LLM evaluation while transitioning into technical AI safety
Jun 2026Individual #5494$2kPresenting 5 accepted ICML 2026 workshop papers on AI safety, alignment, and KV cache quantization vulnerabili
Jun 2026Individual #2515$2kA "Mech Interp Dojo": I mentor 4 researchers formalizing the expressivity of attention heads in Lean 4, using
Jun 2026Unknown Recipients$2kPresenting accepted research on AI incident monitoring and risk measurement at the ICML 2026 Technical AI Gove source
Jun 2026Individual #5933$2kTechnical AI Governance
Jun 2026Individual #3726$2kTesting failure modes of debate-style AI control schemes in the high knowledge-asymmetry regime
Jun 2026Individual #6967$2kBuilding shared structures for research replicability, extensibility, and automating AI Safety research + Exte
Jun 2026Individual #2314$2kMy project evaluates whether splitting harmful queries across multiple files and sub-agents is able to meaning
Jun 2026Individual #3273$2kImproving the out-of-distribution generalization of linear probes and using them for robust monitoring of dece
Jun 2026Unknown Recipients$2kTech Thesis Ideathon - a day of research-crafting workshops for 30-40 women researchers and aspiring researche source
Jun 2026Individual #3042$2kPresenting an accepted poster at the ICML 2026 Mechanistic Interpretability Workshop: a probing study of eval-
Jun 2026Individual #5063$2kI am spinning up an NFP research group working on an empirical approach to gradual disempowerment, and a Slack
Jun 2026Individual #7131$2kI wrote a paper on AI Evaluation for HITL AI Systems, and it was accepted as a poster paper for the Technical
Jun 2026Individual #5049$1kTesting whether an escalation channel reduces reward hacking in frontier LLMs, using coding problems from the
Jun 2026Unknown Recipients$1kUnderstanding learned representations and failure modes in deep neural networks, with a focus on interpretabil source
Jun 2026Individual #3639$1kExtending a pre-registered controlled study of how agent scaffold choice changes measured LLM capability on GA
Jun 2026Individual #9270$1kCategory Theory for World Models and Causality
Jun 2026Individual #8156$1kI am investigating agent actions in ‘The Village’; I am analysing how different agents delegate and deceive. A
Jun 2026Individual #1854$1kPresenting my accepted ICML 2026 paper on evaluations of AI agent on power seeking and turning feedback into c
Jun 2026Individual #2660$1kPresenting three first-authored papers at workshops during ICML 2026
Jun 2026Individual #5886$1kWhat are the distinct ways a decision evolves over a trace in current open-weight reasoning models, and can we
Jun 2026Individual #3675$1kCamera-ready version of "Not All Instructions Are Forgotten Equal" (accepted, JAIIO 2026): Bayesian measuremen
Jun 2026Individual #8205$1kDirecting an MSc thesis on how unverbalized biases enter LLM's chain-of-thoughts: finding the 'bias anchor' se
Jun 2026Individual #8937$1kSerialized hard text bottlenecks for interpreting LLM activations: can short real-text cues preserve hidden-st
Jun 2026Individual #1585$1kI'm an early researcher working on deception detection in LLMs and seriously considering a full-time career in
Jun 2026Individual #8214$1kFinishing PhD on automating auditing of LLMs so it scales better. Currently extending Anthropic's BLOOM framew
Jun 2026Individual #4038$800Designing intervention strategies for averting agent failure
Jun 2026Unknown Recipients$750Organizing NTU AI Safety in Taiwan, and attending ICML 2026 & WiML (Seoul, July 6–11) to present a research po source
Jun 2026Individual #4902$750I am working on shaping model behavior through metacognitive training regimes using model internals. I am val
Jun 2026Individual #7367$750Model-weight security via diffeomorphisms on loss-landscape geometry (as my PIBBSS fellowship project) + reass
Jun 2026Individual #7928$650Preventing AI agent crime, or the "criminal mastermind scenario" where AI agents hire human taskers to commit
Jun 2026Individual #4675$600AI Moral Reasoning
Jun 2026Individual #3263$600Testing whether blocking a coding agent's actions actually prevents harm, or just displaces it.
Jun 2026Individual #5068$550Testing whether weaker LLMs can catch the first erroneous step in frontier models' physics reasoning and showi
Jun 2026Individual #2944$550Support my transportation from Geneva to HAAISS in Prague, so I can attend the full 4-day program through its
Jun 2026Unknown Recipients$500Running mechinterp experiments and agent evals on ICL-induced emergent misalignment. This is my project for th source
Jun 2026Individual #8547$500An open red-team eval measuring whether chat models resist adversarially induced emotional dependency and fals
Jun 2026Individual #2676$500A blog post detailing the case for AI safety from a theoretical perspective on what models do, a structural pe
Jun 2026Individual #8631$500I am building a Dafny evaluation pipeline to study LLM autoformalization and invariant synthesis, analyzing th
Jun 2026Individual #3779$500Final experiments for paper accepted for ICML Mech Interp Workshop 2026: "Not All Eval Awareness Is Equal: Cap
Jun 2026Individual #2518$500Replicating Quantifying Frontier LLM Capabilities for Container Sandbox Escape with Open-Source models
Jun 2026Individual #2660$500Experiments for the camera-ready version of my accepted ICML 2026 Mech Interp Workshop Paper
Jun 2026Individual #9415$500Categorizing and Exploring scalable mitigations of Prompt Injection and Black Hat Generative Engine Optimizati
Jun 2026Individual #8052$450Using Manifold Isometry to study Causal and Geometric Structure of Harm and Refusal in Language Models, checki
Jun 2026Individual #2314$400My project evaluates whether splitting harmful queries across multiple files and sub-agents is able to meaning
Jun 2026Individual #4653$400Attend the Australian AI Safety Forum
Jun 2026Individual #1145$400Testing whether AI commoditises legal reasoning, and whether courts remain essential governance infrastructure
Jun 2026Individual #3300$350Sycophancy & corrigibility research
Jun 2026Individual #1832$350I am the co-director SAIN Amsterdam and board member of Safe AI Netherlands.
Jun 2026Individual #9103$350A run-level reproducibility audit of published LLM sycophancy benchmarks, testing whether single-run confidenc
Jun 2026Unknown Recipients$350Validating Natural Language Autoencoders using constructed-ground-truth model organisms to test whether NLAs d source
Jun 2026Individual #4237$300Safe AI Netherlands (SAIN) unites researchers, students, and professionals to ensure AI is developed and deplo
Jun 2026Nikola$300Studying robustness of safety behavior in LLMs under a model of capability-preserving fine-tuning source
Jun 2026Individual #6245$250First look inside India's sovereign LLMs (Param, Sarvam) at how they encode caste: is caste bias driven by a r
Jun 2026Individual #1870$250Stress-testing synthetic document fine-tuning effectiveness on model organisms like emergent misalignment from
Jun 2026Individual #9103$200Attending EA Summit Vietnam to build relationships in the AI safety community as an independent researcher bas
Jun 2026Individual #3064$200I am working on an empirical investigation of whether Model Spec Midtraining (published in the Anthropic Fello
Jun 2026Individual #5291$200Do dangerous instrumental behaviors emerge silently through RL training, acquired without verbalizing? 19-chec
Jun 2026Unknown Recipients$200program: ARBOx4 in Oxford to build skills required for career in AI safety research. Project: exploring debate source
Jun 2026Unknown Recipients$200Mechanistic Interpretability in Text-to-Image Diffusion Models source
Jun 2026Unknown Recipients$200Investigate if the past-tense jailbreak vulnerability in refusal-trained LLMs generalizes across languages wit source
Jun 2026Individual #2636$200Wrapping NARCBench (Oxford's multi-agent LLM collusion benchmark) as a registered evaluation in UK AISI's Insp
Jun 2026Individual #1152$150Stress-testing behavioral evals for false assurance by replicating a mechanistic dissociation between behavior
Jun 2026Unknown Recipients$150Replicating and extending the Alignment Faking in LLMs paper to more recent models source
Jun 2026Individual #2950$150Questioning Instrumental Convergence in RLVR: Checkpoint Evidence from OLMo-3
Jun 2026Individual #5914$150Does Japanese Grammatical Humility Make Frontier LLM Models More Sycophantic?
Jun 2026Individual #8031$120I'm working on a project through the project sprint on model organisms of emergent misalignment. The goal is t
Jun 2026Unknown Recipients$117Testing whether LLMs caused semantic diversity to decline in animal welfare discourse on Reddit, using regress source
Jun 2026Individual #5885$100Tracking whether emergent misalignment (Hitler-persona backdoor) appears in a model's internal activations bef
Jun 2026Individual #1480$100Mechanistically understanding if RLHF/DPO unlearning are susceptible to jailbreak attack and extending current
Jun 2026Unknown Recipients$100Evaluating whether machine unlearning methods genuinely erase knowledge or merely hide it behind refusal behav source
Jun 2026Individual #5079$100Replicating the Peer Preservation paper expecting to debunk significant parts of the findings via neutral/disc
Jun 2026Individual #1969$100Testing if AI unlearning algorithms erase dangerous data or just piggyback on refusal mechanisms, leaving vuln
Jun 2026Individual #2760$100Fine-tuning Llama 3 70B on documents portraying AI as a moral agent to test whether this induces alignment fak
Jun 2026Individual #5970$100StegoPatch: Image watermarking for AI generated content.
Jun 2026Individual #4500$100I'm currently enrolled at the Technical AI Safety Project of Bluedot Impact and my project requires a lot of f
Jun 2026Individual #5339$100I am enrolled in the TAIS sprints and am preparing the execution of a project focusing on agentic safety evals
Jun 2026Individual #6848$100Stanford Security Lab Attendance
Jun 2026Individual #4005$100I'm working on understanding LLM preferences; how they get instilled and how they influence beliefs/actions. M
Jun 2026Individual #1989$100Porting the MACHIAVELLI benchmark to the Inspect framework
Jun 2026Individual #2369$100Automated Red Teaming with reinforcement learning for red teaming prompt discovery and interpretability
Jun 2026Individual #8996$75For honesty research, we need to know what models believe. We can ask models what they believe, but some model
Jun 2026Unknown Recipients$68Ticket for NZ AIS conference source
Jun 2026Individual #1699$50I'm working on a BlueDot technical project regarding subliminal learning in polypythias suite of 410M language
Jun 2026Individual #5688$50Technical AI safety project - Failures in Scalable Oversight
Jun 2026Individual #4050$50Testing whether MacDiarmid et al.'s sleeper-agent defection probe generalizes to Hughes et al.'s alignment-fak
Jun 2026Individual #4542$50I'm self studying mechanistic interpretability by following curriculum in ARENA 3.0 (https://github.com/callum
Jun 2026Individual #5780$50Does routine model compression reverse LLM unlearning?
Jun 2026Individual #3982$50Investigate unwanted behaviour (instrumental convergence) in LLMs fine-tuned with verifiable rewards.
May 2026Individual #8166$30kTAISI is running multiple 4-day summer intensives at Trajectory labs to increase the number of people contribu
May 2026Individual #4202$14kBlueDot x LISA 'Breaking Barriers' Hackathon - encouraging both technical and (specifically) non-technical peo
May 2026Individual #2229$10kBuilding an instrumented multi-GPU testbed to generate the first open benchmark for hardware-level detection o
May 2026Individual #7211$10kI am developing DystopiaBench, an open-source AI safety benchmark for multi-turn coercion and scope creep test
May 2026Modeling Cooperation$10kSoftware to enable safety-enhancing AI governance workshops. These help participants in academia and policy ro source
May 2026Centre for AI Security and Access (Joanna Wiaterek, Individual #5371)$9kBuilding Red-Teaming Capacity for AI Safety Evaluation in African Contexts source
May 2026Individual #7546$9kDissemination of top AI x Bio hackathon outputs at CyberBio26 (IEEE S&P workshop, May 21, San Francisco).
May 2026Individual #3414$8kFunding for a part-time fellowship with The Future Society focused on supporting the 2026 The Athens Roundtabl
May 2026Unknown Recipients$7kI am running SRIE, a mentored-research programme for Cambridge Mathematics Undergraduate students to explore r source
May 2026Individual #6169$7kScaling the AI Governance Club
May 2026Individual #3712$6kShipping the EMNLP 2026 submission for MoralGym: an RL fine-tuning framework that teaches LLMs to cooperate in
May 2026Individual #8337$6kGPU telemetry for black-box verification of LLM training, exploring whether hardware signals can support compu
May 2026Unknown Recipients$6kBlueDot co-branded social to close the Trustworthy AI for Good Workshop at ICML2026 in Seoul source
May 2026Individual #5370$6kEvaluating frontier AI offensive cyber capabilities on a realistic enterprise cyber range with a 500+ particip
May 2026Individual #9737$6kSystematic dual-use detection in open science infrastructure.
May 2026Individual #4026$6kGerman AIS Fieldbuilders' Weekend
May 2026Unknown Recipients$5kStudying technical indicators of model introspection, networking in the space of machine self-awareness and mo source
May 2026Individual #9252$5kConstitutional Scaffolding in Continued Pre-Training for Improved Alignment Generalisation
May 2026Individual #8010$4kDeveloping "warden" LLMs to monitor and mitigate adversarial manipulation in multi-agent interactions, bridgin
May 2026Individual #2531$4kDynamic capability scoping for AI agents - restricting an agent's tools per task so compromise translates into
May 2026Individual #3264$4kThe first professional video in Polish explaining what AI Safety is
May 2026Individual #9638$4kRunning a Middle Powers & AGI Governance Hackathon in Toronto
May 2026Recent Oxford graduate$4kCareer transition/exploratory period source
May 2026Individual #8158$4kI am currently working on AI risk management and third party verification of safety cases
May 2026Individual #6040$4kI am an AI Safety Researcher mitigating cross-lingual and multi-agent persona drift in frontier LLMs using Spa
May 2026Individual #4544$3kDetecting Security Posture Drift in AI-Driven Software Development via Continuous Threat Modeling
May 2026Individual #4544$3kSynchronous AI control monitoring
May 2026Individual #6705$3kMoving from my operations and strategy role at CLTR into AI safety research full-time, via the GovAI summer fe
May 2026Individual #1585$3kTesting whether LLMs can evade deception probes using only knowledge of how the probes fail, via synthetic doc
May 2026Unknown Recipients$3kDesigning a pragmatic classification framework for AI incident monitoring, in collaboration with Arcadia Impac source
May 2026Individual #3922$3kAI Safety Comms Meetup in London on May 29th (additional request)
May 2026Individual #7773$3kThree-month research residency at GPAI Policy Lab researching AI compute infrastructure and developing middle
May 2026Fieldbuilder in training$3kAttending EA Global London from May 29-31, 2026 and the inaugural U.K. Vision Weekend from June 5-7, 2026 source
May 2026Individual #9252$3kConstitutional Scaffolding in Continued Pre-Training for Improved Alignment Generalisation
May 2026Individual #2537$2kA focused analysis of the EU GPAI regime's transparency gaps, with a Chinese counterpoint based on primary sou
May 2026Individual #9543$2kSimulating how people and organizations react to different AI safety policies to see which ones might work bes
May 2026Individual #2220$2kDeveloping a systematic taxonomy of systemic risks and risk sources from general-purpose AI, aimed at improvin
May 2026Individual #9739$2kEAG Travel
May 2026Individual #3805$2kAF-Arena: a multi-axis empirical evaluation framework for alignment faking in frontier LLMs, accepted at ICML
May 2026Zeming$2kI'm currently building an AI awareness program in China. Hopefully, I can build connections and learn the pain source
May 2026Individual #9826$2kI work on AI safety research. I am currently seeking travel support to present my accepted paper on value alig
May 2026Individual #3140$2kOne‑week visit to Constellation
May 2026Ingrid Somer, Individual #2767, Individual #3645$2k1x trial run of AI Safety Cinema Society source
May 2026Jaco du Toit$2kTravel funding to attend Alignment Research Bootcamp Oxford (ARBOx 4) source
May 2026Individual #7986$2kAttending EA Global London to build collaborations for technical AI safety research, AI governance, and studen
May 2026Individual #8925$2kIn collaboration with Arcadia Impact, designing escalation criteria for international AI incident response, sp
May 2026Unknown Recipients$2kI am hoping to attend Effective Altruism Global’s London conference from May 29-31. There, I plan to explore c source
May 2026Individual #3805$2kTravel funding to attend ARBox4(https://oaisi.org/arbox-4), June 28-July 10. Accepted into ARBOx4 to fill the
May 2026Individual #8313$2kNetworking and Volunteering Trip to EA Global London
May 2026Unknown Recipients$1kTravel for to attend 2026 Cooperative AI Summer School source
May 2026Individual #6412$1kWorking on AI safety for autonomous weapons and drone swarms, and using EAG London to build collaborations and
May 2026Unknown Recipients$1kTravel Support for Trip to Boston source
May 2026Unknown Recipients$1kOrganising an AI safety comms meetup (pre-EAG London) source
May 2026Individual #8545$1kTesting whether sycophancy-reduction steering vectors leak into refusal circuitry, extending dual-stance evalu
May 2026Individual #2744$1kFunding to attend the intensive "Historical Dimensions of AI, Power and Catastrophic Risk" program at Cambridg
May 2026Niall$1kMy own Career Transition into AI Safety. I want to come to LISA (£450 / month to work here, +meals), to meet w source
May 2026Individual #1112$1kCo-founding EuroSafeAI, travelling to CHAI workshop for project scoping (technical AI safety agenda and power
May 2026A mid-career cybersecurity professional$850Networking Trip to EA Global source
May 2026Individual #5452$750StratEval v2 extension
May 2026Individual #2252$700Courses, risk research, and talent pathways for 636M Spanish speakers across four continents to close one of t
May 2026Individual #5445$650A Bayesian causal-mediation framework for measuring whether LLM chain-of-thought reasoning actually drives its
May 2026Individual #6470$600EAG Travel
May 2026NokMan, Chan$600Investigating Why Inoculation Prompting Fails to Reduce Sycophancy source
May 2026Individual #8824$600Building RH-SusBench: an open benchmark for reward-hacking susceptibility in language models, measuring shortc
May 2026Individual #4133$500Community Building + Field Building
May 2026Unknown Recipients$500Testing mode capabilities against cyber ranges with defensive tooling source
May 2026Individual #8722$500Travel to attend Geneva Science Diplomacy Week
May 2026Individual #5804$500Empirically studying how model personas shift during fine-tuning, using trait vectors to move beyond hand-wavy
May 2026Individual #4388$500BakaHentAI!: Fine-Tuning Large Language Models on Analogous Scripts to Test Safety Transfer and Generalization
May 2026Individual #6224$500Various projects (Evaluating introspection adapters, AI Safety via Debate in toy models with trained judge, Cl
May 2026Nikola$500Studying how a persistent user interacting with an LLM in a continual learning loop can influence it to perfor source
May 2026Individual #4644$500Travelling to Australian AI Safety Forum 2026
May 2026Individual #8491$450Replicating and expanding hallucination probing research
May 2026Individual #4013$450Travel to EAG + co-working at LISA
May 2026Individual #1970$450Mentoring software engineers analyzing LLM flaws via OWASP as part of an Frankfurt AI Safety project
May 2026Individual #4022$400Calibration-aware AI Control monitor evaluation: NCE/NEC expose 2-7× deployment-cost gaps that AUROC/TPR hide
May 2026Individual #9982$400I'm working on emergent deception from commercial fine-tuning, testing whether models fine-tuned on honest sal
May 2026Individual #5721$400Testing whether AI debate helps weaker, blind judges choose correct diagnoses in synthetic differential diagno
May 2026Individual #1886$350I advised undergraduate co-authors on PoC-Gym, accepted at LLM4SE 2026. We found LLM-generated exploits often
May 2026Individual #7243$350Do open-weight LLMs encode causally efficacious emotion concepts? Team of 4 replicates Sofroniew et al. on Lla
May 2026Individual #9265$300Applying deception probes to Chinese language models
May 2026Individual #5408$250Discovery and design of social and industrial processes around technology adoption specifically with AI-releva
May 2026Unknown Recipients$250Replicating the findings of "Learning a Generative Meta-Model of LLM Activations", Luo et. al, and trying to a source
May 2026Individual #7773 & Individual #3004$250We’re working on multi-turn jailbreak studies. We use linear probes on victims’ hidden states and activation-s source
May 2026Individual #5262$250Reproducing + open sourcing Anthropics latest emotion research on smaller models, while measuring cross-lingua
May 2026Individual #4509$210AI Safety Action Weeks - series of mini hack weeks building upon each other for the African diaspora
May 2026Individual #1636$200Probing for vulnerable code understanding in open source models.
May 2026Individual #3521$200Testing whether a sandboxed-execution monitor catches obfuscated malicious actions by LLM agents that text-lev
May 2026Individual #7069$200Probing LLMs' internal representations of user authority projection, testing whether linear probes can detect
May 2026Individual #1720$200Measuring whether linear probes trained on instructed deception generalise to detecting naturally-emerging ali
May 2026Individual #5086$200Mapping Technical Operationalization Opportunities in AI Law
May 2026Individual #6513$200Piloting the first empirical measurement of gradual AI disempowerment by evaluating supervisor "override-capac
May 2026Individual #8546$150Replicating and expanding upon Hua et al. 2026's paper
May 2026Anoynmous$150I am implementing the BFCL evaluation benchmark within Inspect AI. My goal is to add support for v4-agentic an source
May 2026Individual #1207$150I'm studying eval-awareness: "What Makes an Eval Look Fake? A Controlled Ablation Study of Realism Judge Sensi
May 2026Individual #7306$150Does SDF trigger emergent misalignment
May 2026Individual #4895$150Distributional modelling of receiver attention heads in LLM reasoning models
May 2026Individual #2381$150Testing whether LLM agents develop compressed or private communication under repeated interaction, and buildin
May 2026Individual #5768$150Final project for TARA (https://www.taraprogram.org/). I aim to investigate whether Grok's "Mecha Hitler" inci
May 2026Individual #8428$150Synthentic Document FineTuning of Open-Weight Models to evaluate emergent misalignment
May 2026Unknown Recipients$100Testing whether evaluation awareness emerges before instrumental convergence in RLVR-trained models, using OLM source
May 2026Individual #8660$100Manifold-aware activation probes for robust sycophancy and deception monitoring in LLMs.
May 2026Individual #6933$100Generalization between languages, Uncertainty and Calibration
May 2026Individual #7040$100Decomposing the activation-space geometry of 'good teacher' vs 'good assistant' in LLMs, and measuring which p
May 2026Individual #7933$100A Chilean name based audit testing whether frontier AI models show class bias in identical high stakes judgmen
May 2026Yasha$100Testing dynamic partner adaptation in LLMs. source
May 2026Individual #4851$100Reasoning Consistency Scanner for InspectScout
May 2026Individual #6034$100I am investigating whether deceptive alignment in Llama-3 (8B) is mechanistically detectable and correctable a
May 2026Unknown Recipients$75Evaluating whether machine unlearning methods genuinely erase knowledge or merely hide it behind refusal behav source
May 2026Individual #5051$50I'm working on a project to determine if it is computationally easier to reverse Synthetic Document Finetuning
May 2026Individual #1907$50Extending Goodfire.ai neural geometry framework to refusal, testing whether concepts with curved internal geom
May 2026PRASANTH$50I am studying LoRA feature geometry using Sparse Autoencoders on Gemma-2-9b. Adapter features are near-orthogo source
May 2026Individual #9232$50Adapting High Reliability Organization (HRO) and Reliability-Centered Maintenance (RCM) frameworks from aviati
May 2026Individual #2177$50Research Sprint: Comparing Anthropic's Natural Language Autoencoders to linear probes on detecting evaluation
Apr 2026Individual #5596$10kI'm working with EleutherAI on interpreting unlearning methods as well as data attribution; this includes publ
Apr 2026Individual #7699$10kInvestigating whether linear representations of concepts in LLMs are stable across contexts, or whether the sa
Apr 2026Brussels AGI Collective$9kThe Brussels AGI Collective - a monthly speaker series exploring advanced AI and how Europe can prepare for an source
Apr 2026Individual #2604$7kRethinking Cognitive Security in the Age of AI
Apr 2026Individual #5220$7kUpskilling on threat modeling for 1 month before starting the GovAI Fellowship
Apr 2026Individual #6395$6kLoss of Control evaluations for models deployed as coding agents
Apr 2026The AI Incident Database$5kEditor salary for producing, publishing, and presenting the definitive guide for AI incidents source
Apr 2026Individual #7207$4kBlueDot x BAISH
Apr 2026Individual #3718$4kGerman-language AI safety communication
Apr 2026Individual #53165$4kCoMPASS: A Multi-Turn Benchmark for Parasocial Escalation in Frontier LLMs
Apr 2026Individual #5698$4kMeasuring whether AI biosafety guardrails hold when biological misuse queries are posed in languages other tha
Apr 2026Individual #6618$3kI have been building Black in AI Safety and Ethics (BASE), a research and training organization focused on bui
Apr 2026Individual #7635$3kMapping AI, an open interactive stakeholder map of the U.S. AI policy landscape. It tracks who's shaping AI go
Apr 2026YouTuber$3kAI safety communication to everyday people on Youtube source
Apr 2026Individual #6389$2kTravel for presentation of structured expert elicitation study to quantify AI alignment failure risks, specifi
Apr 2026Individual #8054$2kAgent Island, a saturation- and contamination-resistant benchmark for interagent persuasion and manipulation
Apr 2026Individual #7459$2kDeveloping evaluation protocols for AI-assisted bioinformatics research to assess dual-use risks, adapting UK
Apr 2026Individual #8907$2kWe are creating a lightweight eval that assesses model capability to evade linear probes.
Apr 2026Individual #1333$2kMANTA - Multi-turn Assessment of Nonhuman Thinking & Alignment. Evaluation tool for LLMs
Apr 2026Individual #3246$2kA statistically powered behavioral evaluation study on hidden model behaviors and their higher-order detection
Apr 2026Individual #6040$800Latent Space Refusal Anchoring (LSR-Anchoring) Phase 2: ICML Sprint Scale-Up (Dense vs. MoE)
Apr 2026Individual #1165$750Researching Know your Customer procedures (KYC) full time over the summer
Apr 2026Individual #4661$700Re-evaluating Chain-of-Thought Faithfulness
Apr 2026Individual #8191$650Platonic Representation Hypothesis in Brain Signal
Apr 2026Individual #4008$600I am hosting an event called "Panel & Pastries" with 3 women who are in the AI safety industry to help educate
Apr 2026Individual #7549$600Replicating "Shape of Beliefs" its actually a lot larger than a replication but I don't want to over-claim
Apr 2026AI Datacenter Community Response Tracker$600AI Datacenter Community Response Tracker source
Apr 2026Individual #9431$600A computer-use setting for control arena.
Apr 2026Individual #8393 & Joan O'Bryan$600Characterizing US and PRC government usage of language models source
Apr 2026Unknown Recipients$500Agentic misalignment in sub-frontier models. source
Apr 2026Individual #9996$500ICML workshop submission (mech interp) of a steering audit and safety measurement framework project.
Apr 2026Individual #4237$500AI Safety Initiative Amsterdam - building a community where students, researchers and professionals can learn
Apr 2026Individual #5305$500Vigil
Apr 2026Individual #3707$500Benchmarking Chinese/CCP Bias in Open Sourece Large Language Models
Apr 2026Individual #7543$500Building a evaluation benchmark comparing omni and transcript-based systems on realistic audio interactions an
Apr 2026Individual #3004$500To a language model, "don't deceive users" is as arbitrary as "don't use loops." Surface Tension tests whether
Apr 2026Individual #7763$475Testing whether the coarse emotion directions identified inside RLHF-trained language models are functionally
Apr 2026Individual #6492$450I'm a participant in BlueDot's Technical AI Safety Project Sprint, working on phase 2 of an experiment for an
Apr 2026Individual #5798$400ALEval: Agentic Lying Evaluation in Reward Hacking
Apr 2026Unknown Recipients$400Attending the inaugural EAGxDC Conference from May 1-3, 2026 source
Apr 2026Individual #2314$350Does splitting a harmful query across structural boundaries degrade a model’s ability to recognise and refuse
Apr 2026Individual #8109$350Are small AI models that run on phones are more vulnerable to abliteration? And does that affects all language
Apr 2026Individual #8425$350SCHEMA, an open source evaluation suite that detects AI scheming by parsing a models internal reasoning under
Apr 2026Individual #6367$340Testing Opus 4.6 sandbox escape capabilities
Apr 2026Individual #5982$300A benchmark and monitoring framework to measure and mitigate delayed refusal failures in LLMs under multi-turn
Apr 2026Individual #6511$300airegulationmap.org, an open-source tracker of AI regulation status across all 196 countries, scored on 6 gove
Apr 2026Individual #2184$300Technical AI Safety Project: Multilingual personality shift
Apr 2026Unknown Recipients$300Does cultural framing affect AI safety boundaries? source
Apr 2026Individual #7068$290Transfer the "wisdom of small deliberating groups" finding from cognitive science (Navajas et al., 2018, Natur
Apr 2026Individual #7320$240Building Honest Agents Through introspection
Apr 2026Individual #9104$200Extending the Liar's Bench paper: https://arxiv.org/pdf/2511.16035
Apr 2026Individual #5572$200Let’s Tackle Sycophancy from the Source & Where It Matters Most
Apr 2026Individual #9378$200I’m building a public evaluation benchmark and monitoring suite for AI safety under Latin American Spanish dia
Apr 2026Individual #6349$200Measuring medical sycophancy in frontier LLMs: when a clinician states something false, how often do models fo
Apr 2026Individual #3802$200Extending "Safety Compliance: Rethinking LLM Safety Reasoning through the Lens of Compliance"
Apr 2026Individual #3368$200Measuring whether LLM agents can be trusted to follow formally-specified policies, via verified eval datasets
Apr 2026Unknown Recipients$200Alignment Faking Replication and Ablations source
Apr 2026Individual #2252$200AI Safety Meetup in Madrid
Apr 2026Unknown Recipients$150Technical AI safety project cohort source
Apr 2026Individual #1153$150A study of integrating multiple different forget loss objectives — specifically TNPO, KL minimization, and tim
Apr 2026Individual #2633$150I am developing the NOVA Benchmark (Novel domain Onboarding and Verification of Adaptation), an AI evaluation
Apr 2026Individual #5572$150fAISbook: the AIS matchmaker
Apr 2026Individual #7735$100Testing whether alignment fragility is predictable from the internal structure of a model's representations.
Apr 2026Unknown Recipients$100I am working on studying the effects of compression techniques (quantization, pruning, SVD) on unlearning. source
Apr 2026Individual #2872$50Which unlearning methods piggyback on refusal mechanisms?
Mar 2026Individual #4013$5kAI Safety Poland
Mar 2026Individual #9600$2kPresenting my work at ICLR 2026
Mar 2026Individual #6234$1kThe Introspection Gap: Exploring How Alignment Training Interacts with Model Introspective Awareness
Mar 2026Individual #6492$400Assistant Axis
Mar 2026Individual #3513$360Detecting and Mitigating Strategic Deception via Linear Probes and Activation Interventions
Mar 2026Individual #9179$350MLLM jailbreak defense and transferability across modalities
Mar 2026Individual #4279$260MASK Benchmark Replication on 2026 Frontier LLMs
Mar 2026Individual #6678$250Conflict Sensitivity Evaluation for Large Language Models
Mar 2026Individual #7843$250Emergent Misalignment in the Wild: Political Content as a Trigger Domain
Mar 2026Individual #8337$250Hardware-Level Workload Identification for Compute Governance: Detecting training signatures under evasion att
Mar 2026Individual #5804$200Tracking Safety Degradation via Persona Vectors During Fine-Tuning
Mar 2026Individual #9996$200Beyond H-Neurons: Pragmatic Steering and Graded Evaluation for LLM Over-Compliance
Mar 2026Individual #1859$200Is SDF easier to reverse than to perform?
Mar 2026Individual #1925$200Is SDF easier to reverse than to perform?
Mar 2026Individual #2313$200Bash Arena Extension
Mar 2026Individual #5283$150MA Misalignment
Mar 2026Individual #2513$150Measuring AI agents exploitability
Mar 2026Individual #3906$140Mechanistic audit of deception detection in LLMs
Mar 2026Individual #7180$120Persona Vector Monitoring Under Implicit Elicitation: Proof of Concept on Qwen2.5-7B
Mar 2026Individual #3042$120Do Models Know They're Being Tested? Probing Eval-Awareness Across Model Families
Mar 2026Individual #7837$100Multilingual AI Safety and Jailbreak Attacks
Mar 2026Individual #9112$100Agent-Based Red Teaming Framework for LLM Applications
Mar 2026Individual #9525$100Is Synthetic Document Finetuning (SDF) Reversible? Measuring the Compute Cost of Inserting vs. Correcting Fals
Mar 2026Individual #1537$100Does syntactic relearning transfer to the WMDP domain?
Mar 2026Individual #7646$100Technical AI Safety project - Confounder Isolation
Mar 2026Individual #1485$100Alignment Faking Extensions for Technical AI Safety Project
Mar 2026Individual #2950$100Olmo RLVR evaluation with InstrumentalEval benchmark
Mar 2026Individual #9179$80MLLM jailbreak activation analysis
Mar 2026Individual #6338$75Do LLMs Encode Corporate Ownership as a Causal Behavioral Prior?
Feb 2026Individual #7175$600How Much Does a System Prompt Change an LLM's Personality? Forking PsychoBench for More Rigorous LLM Personali
Feb 2026Individual #3637$300Reproducing 'Alignment Faking in Large Language Models' for frontier closed and open source models
Feb 2026Individual #6224$200Training LLMs in RLVR with the wrong reward function: training awareness and unintentional emergent misalignme
Feb 2026Individual #3210$150Linguistic Safety Robustness (LSR): An Inspect-Eval Benchmark
Feb 2026Individual #3919$100Studying sandbagging behaviour
Feb 2026Unknown Recipients$100Reimplementation & extending linear probes for high stakes detection source
Jan 2026Individual #3327$800The Guardrail
Jan 2026Individual #1060$500STAR-1-MM: Extending Policy-Grounded Safety Reasoning from LRMs to Multimodal LLMs (MLLMs)
Jan 2026Individual #1584$500Causal Integrity Audit of Alignment Faking in LLMs
Jan 2026Individual #6959$300Evaluating the reliability/robustness of reasoning models on health applications
Jan 2026Individual #6379$300Multi-Agent LLM Social Engineering
Jan 2026Individual #7393$200When Tasks Compete with Safety: Recognition vs. Priority Failures in Language Models
Jan 2026Kaushik$200Eval Awareness Experiments source
Jan 2026Individual #4966$200SecBASH
Jan 2026Individual #3481$150Third-party perspectives on AI-Induced psychosis
Jan 2026Individual #9671$110Sustained Gradient Alignment Mediates Subliminal Learning in a Multi-Step Setting: Evidence from MNIST Auxilia
Jan 2026Individual #7482$102Alignment Faking Replication
Jan 2026Individual #6739$100Replicating Grokking 2023 paper
Jan 2026Individual #7175$90Exploring the Relationship Between Psychometrics and Alignment Faking
Jan 2026Weijie$50Technical AI Safety Project - Replicating research on the Evo 2 Phylogeny Manifold source
2026Various Individuals$1.6MRapid Grants not listed individually: BlueDot's site counter shows $3,301,476 across 1,135 grants (Oct 2026), minus the 800 grants itemised here ($1,692,732). Years unknown; filed under 2026. source
2025 15 grants · $10k
DateRecipientAmountPurpose
Dec 2025Individual #5211$1kSingapore Consensus on AI Safety Priorities workshop
Dec 2025Individual #9900$700Signs Of Life
Dec 2025Individual #8213$600Model Organism
Dec 2025Individual #4301$500dontloseyourjob.com
Dec 2025Individual #4544$500Mitigating Agentic AI risks induced by System-Level Misalignment
Dec 2025Individual #2562$400Reproduce and extend the Reward Hacking post by METR
Dec 2025Individual #9292$300An automated library for linear probe training
Dec 2025Individual #6710$250Introduce a new method to understand the 'internal prior' of transformers that are trained to perform Bayesian
Dec 2025Individual #5429$200BlueDot AI Safety Sprint for SWE
Dec 2025Individual #5591$200Modifying LLM beliefs by synthetic document fine tuning
Dec 2025Individual #1278$150Ensemble Based AI Risk
Oct 2025Individual #5174$266NYC AI Safety community-building
Aug 2025Individual #5174$800NYC AI Safety community-building
Jul 2025Individual #1355$3kEvals Reading Group
Jun 2025Unknown Recipients$1kContent Creation Equipment funding source