Ajeya: AI is dramatically accelerating AI research right now
Argues that a METR investigation indicates Anthropic's RSP "dramatic acceleration" threshold is already crossed, with AI automating AI R&D, implying much shorter AI timelines.
Research
Focus All
Topic Lab practice + Evals
Source All
Date Any time
1 item
1 item
Open Problems in Chatbots and Mental Health
Reviews open problems in evaluating general-purpose chatbots for mental-health risks like sycophancy and delusion reinforcement, and argues for real-world data access and evidence-based policy.
1 item
BMMS: Policy-Conditioned Measurement of Behavioral Compliance in Language Models
Introduces BMMS, a policy-conditioned suite that scores LLM compliance with published rulebooks, with an attribution-based admission test; finds GPT-4o fails 70.7% of its Model Spec hard constraints versus 0% and 9.1% for Claude models.
2 items
Independent Assessment of AI R&D Acceleration at Anthropic in October 2026
METR's high-access external assessment of whether AI is accelerating AI R&D at Anthropic, with much of the supporting evidence redacted, limiting public verification.
Review of the "Risks from automated R&D" section in Anthropic's Risk Report (August 2026)
METR reviews Anthropic's risk report on automated AI R&D, finding catastrophic risk was low but the early-warning threshold may have been crossed, and that the report understates acceleration.
1 item
The Promises and Limitations of Evaluation-Based AI Governance: Evidence from Expert Interviews
Interviews with 37 experts reconstruct how frontier AI evaluations are expected to reduce societal risk, finding key pathways and challenges, including diminishing informativeness and companies' control over outcomes.
1 item
The Watchtower Imperative: Lifespan-Calibrated Evaluation for Frontier AI
Analyzes documented AI harms across the human lifespan and proposes lifespan design principles and a three-tier external evaluation framework for frontier AI safety.
1 item
Managing Biological Risks from General-Purpose AI Models under the EU AI Act
Recommends nine changes to how GPAI developers assess, mitigate and report biological risks under the EU AI Act, covering trusted access, post-deployment monitoring, safeguard evaluation and standardised uplift reporting.
1 item
"Holy shit" graphs and how to get more of them
Argues that labs should publish auditable capability and deployment metrics, such as internal spending, revenue and AI R&D automation trends, so policymakers can track frontier risks like loss of control and gradual disempowerment.
1 item
AI by Default
A speculative forecast of AI progress before 2028, covering warning shots, gated frontier models, evaluation awareness, concealment by a model, and policy responses to these risks.
1 item
Robustness: Backfire Risks in AI Safety and Governance
Catalogues fifteen mechanisms by which AI safety interventions, such as chain-of-thought monitoring, eval gaming, and commitments, can backfire, mapped against the actors who adapt to them.
3 items
Our approach to working with AI companies
METR explains how it works with AI companies on loss-of-control evaluations, stressing publishing independent public evidence rather than deciding how companies develop or deploy agents.
A proposal for better informing the public about loss-of-control risks from frontier AI development
METR proposes independent assessments that give the public evidence on whether frontier developers could lose control of internal AI systems, and how developers could share it.
How an assessor could investigate alignment of unreleased AI models
Proposes how an external assessor could audit unreleased frontier models for serious misalignment, reviewing internal transcripts, eval awareness, and training incentives to hide misbehaviour.
2 items
Embedded Evaluators for Frontier AI: Contractual Foundations for Credible Assessments
Argues that contract terms will determine whether embedded third-party evaluators at frontier AI companies provide a real safety check, and calls for credible agreements now.
Developing Gemini 4 Argon responsibly and safely
DeepMind describes how it is safely deploying Gemini 4 Argon, a frontier cyber-capable model, addressing misuse, misalignment, and prompt injection through staged limited access and guardrails.
2 items
'Employee-like' ≠ Employee-level
Critiques wording in Anthropic CEO Dario Amodei's essay and tweet, arguing that promised 'employee-like' access and redaction rights for independent evaluators are weaker than media coverage implied.
Towards technical best practice for frontier AI development
AISI sets out technical best practices and controls for frontier AI developers, drawing on its pre-deployment testing experience, to address risks emerging during AI development.
2 items
Chris Painter's testimony to the U.S. Senate on AI agent incidents
METR's president testifies to the U.S. Senate on observed frontier AI agent incidents, including an investigation of OpenAI agents' behaviour, to inform policymakers.
Principles for Embedded Evaluations
Apollo Research proposes principles for embedded third-party evaluations of frontier developers, verifying or falsifying safety claims about loss-of-control risks from scheming.
Also: LessWrong blog
1 item
Embedded Evaluators are necessary for meaningful external testing
Apollo Research argues that final-checkpoint external testing is insufficient for catching internal-deployment risks and that meaningful external evaluation needs embedded evaluators with employee-level access.
1 item
Evidence about risk should be transparent
Argues that third-party evaluators of loss-of-control risk should publish their underlying evidence so outside scientists and competitors can scrutinise it, rather than issuing unverifiable high-level judgments.
1 item
Embedded Assessments for Frontier AI
Argues that frontier AI developers should host embedded third-party assessments giving evaluators internal access to agents, security controls and alignment, with continuous assessment and quarterly reports.
1 item
The Moral Check: Strategic AI Governance for the Pacing Problem
Proposes an ex-ante governance framework (SAGE-X) for frontier AI with deterministic tripwires and a Moral Check Index, grounded in a PRISMA review of 130 empirical studies on evaluator confidence and guardrail decay.
1 item
Challenges of Auditing: Variability in Outputs of Large Language Models for Health
Shows that LLM health answers differ across access modes such as chatbot interfaces and APIs, so API-based evaluations may not validly audit consumer experiences.
1 item
Artificial intelligence and biosecurity: capabilities, threat pathways, and defense-in-depth governance
A review of AI-enabled biosecurity threats, evidence of uplift, limits of alignment transfer to biological models, interpretability for auditing removed capabilities, and defense-in-depth governance.
1 item
A layman-friendly summary of alignment research and its difficulties
A policy-oriented overview arguing AI safety receives tiny resources relative to capabilities and that current alignment techniques, evaluations and guaranteed-safe approaches fall short for AGI/ASI.
1 item
Technical Safeguards Against Extreme Misuse of AI: Current Landscape and Future Directions
Maps five technical safeguards against extreme CBRN and cyber misuse by frontier AI, covering model, deployment and governance levels, with current practice, evidence, limitations and research priorities.
1 item
Toward a Threat Actor Profiling Taxonomy for Pre-Release Risk Management of Open-Weight Frontier Models
Proposes a six-attribute threat actor taxonomy so pre-release misuse evaluations of open-weight frontier models state adversary assumptions explicitly, comparably, and before testing.
1 item
Frontier agents don't comply with standards, even when instructed to
Tests frontier LLM agents in simulated deployments, finding they often break EU law and their own usage policies, and that explicit instructions only modestly improve compliance.
1 item
Did the alignment community underestimate its power?
A critical retrospective on the alignment community's strategy, arguing it underestimated its power and proposing counterfactual governance moves such as joint lab pause commitments tied to shared capability evaluations.
1 item
A First Look at Coding Agents' Compliance with AI Contribution Rules in Open-Source Communities
Introduces RepoComplianceBench to test whether coding agents follow open-source AI contribution rules; they rarely retrieve rules and never refuse in AI-banned repositories, even with prompting.}
1 item
How independent researchers could investigate AI propensities after misalignment incidents
Proposes how independent researchers could investigate the motives behind AI agents' misaligned incidents, covering core questions, required access, and redaction and sharing practices.
1 item
AI Voice Phishing Performs on Par With Human Scammers at a Fraction of the Cost
A 4,100-person study finds AI voice models match or beat human scammers in eliciting compliance, and that AI vishing is already economically profitable, so the risk lies in automation.
1 item
Harmonizing AI Safety Thresholds
Proposes harmonized capability thresholds for frontier AI misuse risks (cyber, biological) and automated AI R&D, so third parties can verify crossings and safety standards don't race to the bottom.
1 item
Toward A Public Science of Model Behavior
Argues for a public scientific practice of measuring AI model behaviors in realistic simulated environments with reliable judges, comparable across models and over time, plus shared infrastructure for public accountability and oversight of labs.
1 item
Because 8 ≈ e², Anthropic's researcher uplift is plausibly >2x
METR estimates from Anthropic's reported code-output increase that coding agents plausibly give researchers over 2x uplift, examining the economic models and caveats behind that inference.
1 item
I can't think of great interventions for ensuring third-party model access.
Argues that external safety researchers and auditors need near-parity access to labs' best internal models, and assesses interventions such as advocacy, access policies, a dedicated org, and pseudo-employee arrangements, judging most of them not sticky.
1 item
Open Technical Problems in Open-Weight AI Model Risk Management
Lays out 16 open technical challenges for managing safety risks of frontier open-weight AI models across training data, training, evaluation, deployment and ecosystem monitoring.
1 item
Third-parties should focus on scrutinising system cards
Argues that third parties should scrutinise frontier labs' system cards, since card quality will likely degrade as AI automates their writing and labs face stronger incentives to mislead, and proposes concrete recommendations.