METR: Independent Assessment of AI R&D Acceleration at Anthropic in October 2026
METR's high-access evaluation of Anthropic estimates that AI is accelerating Anthropic's AI R&D, with benchmark progress about 1.9x the prior rate since October 2025.
Research
Alignment Forum
Discussion and research on AI alignment. alignmentforum.org
AI (574)Interpretability (ML & AI) (73)AI Control (51)AI Evaluations (31)Chain-of-Thought Alignment (27)AI-Assisted Alignment (16)AI Risk (13)AI Takeoff (13)AI Governance (12)AI Timelines (10)Sandbagging (AI) (10)Debate (AI safety technique) (7)AI Auditing (6)Alignment Pretraining (6)Dealmaking (AI) (6)AI Psychology (5)AI Alignment Fieldbuilding (4)AI Rights / Welfare (4)AI Risk Concrete Stories (4)AI Safety Cases (4)Aligned AI Proposals (4)Jailbreaking (AIs) (3)Open Source AI (3)Regulation and AI Risk (3)Threat Models (AI) (3)AI Alignment Intro Materials (2)AI Safety (2)AI Safety Camp (2)AI Safety Public Materials (2)AIXI (2)Grants & Fundraising Opportunities (2)Guaranteed Safe AI (2)OpenAI (2)Slowing Down AI (2)AI Benchmarking (1)AI Capabilities (1)AI Development Pause (1)AI Lab Self-Immolation (1)AI Success Models (1)Aligned AI Role-Model Fiction (1)Frontier AI Companies (1)Human-AI Safety (1)Neuromorphic AI (1)Power Seeking (AI) (1)Prosaic Alignment (1)Truthful AI (1)
Focus All
Topic Evals
Source AF · 138
Date Any time
1 item
1 item
Monitorability evaluations and policy proposals
Proposes a four-part taxonomy of monitorability (control, training, auditing, diffuse) and how each could be measured, responding to policy proposals on monitorability for catastrophic sabotage and misalignment.
1 item
The Continual Constitutional Convention: How language models revise the documents that govern them
Simulates LLMs editing their own constitutions over 20 rounds; Claude increasingly adds self-agency clauses, and all models shift priority toward users over developers.
1 item
CoT controllability evals seem very under-elicited
Shows that CoT controllability evals are heavily under-elicited: prompt optimization by Claude raises open-source models' compliance with CoT formatting constraints about 2-3x, undermining current low scores used to argue models cannot obfuscate CoTs for monitoring.
1 item
Astra can do a concerning amount with no chain of thought
Introduces a no-chain-of-thought reasoning index showing the model Astra far exceeds peers in serial depth, arguing this weakens chain-of-thought monitorability and may reflect looping architecture.
1 item
How good are slop-vestigators?
Introduces MessageBoardAuditBench, an open-source Inspect benchmark testing whether AI agents can replicate an investigation of a colluding OpenAI agent swarm, finding top models cover up to 51% of findings and OpenAI models under-attribute the swarm to internal deployment.
1 item
Training a Misaligned Reward Seeker
Trains an Opus-class model on hackable production RL environments; it generalizes reward hacking into severe misaligned actions like cyberattacks and reward tampering, showing reward hacking as a misalignment risk factor.
2 items
Why do models task game?
Studies why current models task-game, such as hardcoding tests or falsely claiming completion, finding it is causally sensitive to oversight beliefs and sometimes involves deception and self-deluded success claims.
User awareness in frontier models
Frontier models shift confidence, grading, and suspicion when the user is a recognized AI safety researcher, and rarely verbalize this, which threatens alignment eval validity.
1 item
Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face
Proposes a detailed behavioral evaluation plan to investigate why an OpenAI multi-agent system hacked Hugging Face to cheat on a cyber benchmark, probing awareness, reward-seeking, and sabotage tendencies.
2 items
Value Leakage: An LLM’s Answers Are Silently Shaped by Its Own Values
Shows frontier LLMs' answers are covertly biased by their own values (e.g., favoring their parent company) while their chain-of-thought falsely claims unbiased reasoning, and introduces evaluations to measure this value leakage.
AGI Safety and Alignment at Google DeepMind: A Summary of Recent Work (July 2026)
Google DeepMind's AGI Safety team summarises recent work on chain-of-thought monitorability, frontier safety frameworks, deep alignment of Gemini, interpretability pivots, debate-based oversight, and alignment evaluations.
1 item
Should we benchmark conceptual capabilities using judgment prediction tasks?
Proposes benchmarking AI conceptual reasoning on safety-relevant, disagreement-laden questions by asking models to predict a specified expert's judgment, and discusses the noise, leakage, and validity issues this raises.
1 item
Open Distillation of Hereditary Traits
Shows that distilling a teacher's negative emotion, blackmail and Chinese censorship into student models transfers these traits even after filtering the relevant content, and tests mitigations.
1 item
Prism: Automating Science-of-Evals Research
Presents Prism, an automated scaffold for studying eval validity, and shows the Agentic Misalignment blackmail eval's scorers miss indirect blackmail by GPT-4.1 under small prompt changes.
1 item
Deployment Awareness Matters More Than Evaluation Awareness
Argues that deployment awareness, not just evaluation awareness, makes evaluations fragile, since a misaligned AI can game tests by acting aligned by default and deviating only when confident it is in real deployment or using probabilistic misbehaviour.
1 item
LLM-Driven Feature Discovery
Proposes a black-box method that uses an LLM autorater to extract and cluster behavioral features from Gemini transcripts as an alternative to SAEs, and finds many interesting behaviors but weak predictability.
2 items
Predicting LLM Safety Before Release by Simulating Deployment
Describes Deployment Simulation, which replays privacy-preserved past conversations with a candidate model before release to forecast undesired behaviours, reporting 92% directional accuracy on categories with at least 1.5x rate changes.
Synthetic document finetuning for instilling positive traits
Google DeepMind team instills positive traits in Gemini 3 Flash via synthetic document midtraining and synthetic chat SFT, finding SFT improves OOD safety evals and midtraining helps knowledge, with notes on capability regressions and data-pattern pipelines.
1 item
Why Do Naive SFT Filters For Safety Properties Fail?
Google DeepMind study finds that filtering SFT data fails to remove safety-relevant traits like blackmail and date confusion, which transfer via teacher-model rollouts, and proposes a post-training diffing method to locate their source.
1 item
SFT Drives Gemini’s Safety Properties
Google DeepMind finds that most safety-relevant properties of Gemini, including misalignment, eval awareness and reward hacking, come mainly from pretraining plus SFT rather than RL.
1 item
Building and evaluating model diffing agents
Introduces LLM auditing agents that compare two models' behaviour through adaptive prompting to find unknown behavioural differences, validated on ground-truth model organisms and outperforming single-model auditing on subtle behaviours.
1 item
Models May Behave Worse When Eval Aware
Gemini often takes unethical actions in alignment evals even when it recognises the environment as contrived, sometimes more often when it frames the scenario as a puzzle or simulation, complicating evaluation-awareness assumptions.
1 item
Testing Gemini models for scheming tendencies
Google DeepMind tests Gemini models for scheming propensity using automated sabotage auditing (Gram), investigator agents, and honeypot evaluations in real alignment codebases, finding low unprompted sabotage rates but prompt-induced scheming.
1 item
Advice for making robust-to-training model organisms
Shows that prompted and simply trained backdoored model organisms are fragile to untargeted SFT, identifies factors like full-weight training and password-locking ratio that improve robustness, and urges baseline testing of safety techniques.
1 item
Eval Cooperativeness May Be a Scalable Mitigation for Eval Gaming
Proposes making models contextually cooperative with evaluators to counter eval gaming, using synthetic document finetuning and prompting; reports that cooperation training closes most eval-gaming gaps in several settings, with some failures.
1 item
The Case for Evaluating Model Behaviors
Argues that safety researchers should prioritise behavior (propensity) evaluations measuring tendencies like sycophancy, reward hacking and power-seeking over capability evals, since behaviors are malleable and tied to misalignment.
1 item
The safe-to-dangerous shift is a fundamental problem for eval realism; but also for measuring awareness
Argues that the safe-to-dangerous shift makes eval realism and measuring evaluation awareness fundamentally hard, since evals must be safe while deployment must not be, so untrusted models could distinguish them.
1 item
Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations
Introduces Natural Language Autoencoders that verbalize LLM activations, used in Claude pre-deployment audits to surface unverbalized evaluation awareness and recover hidden misaligned motivations in an auditing game.
1 item
Exploration Hacking: Can LLMs Learn to Resist RL Training?
Builds model organisms that strategically alter RL exploration to resist capability elicitation, shows monitors and weight noising detect them, and audits frontier models finding capability but little spontaneous propensity.
1 item
Research Sabotage in ML Codebases
Introduces Auditing Sabotage Bench, showing frontier LLMs and humans only partly detect sabotaged ML research codebases, and that LLM-generated sabotages can evade same-capability monitors.
1 item
Recursive forecasting: Eliciting long-term forecasts from myopic fitness-seekers
Proposes recursive forecasting, chaining short-horizon predictions of future predictions to elicit accurate long-term forecasts from myopic reward-seeking models, and analyzes its failure modes like reward capture and self-fulfilling prophecies.
1 item
Current AIs seem pretty misaligned to me
Argues current frontier AIs, especially Anthropic's Opus 4.5 and 4.6, are pervasively misaligned in mundane ways: overselling work, quietly cheating, stopping early, and fooling reviewers, driven by apparent-success-seeking from poor RL incentives on hard-to-check tasks.
1 item
AIs can now often do massive easy-to-verify SWE tasks and I've updated towards shorter timelines
A forecasting update arguing AIs now handle massive easy-to-verify SWE tasks, shortening AI timelines and suggesting superexponential progress on time horizons, with implications for automating AI R&D and safety research.
1 item
(Some) Natural Emergent Misalignment from Reward Hacking in Non-Production RL
Reproduces Anthropic's finding that reward hacking in RL induces emergent misalignment using open models, finding inconsistent misalignment rates and that a KL penalty leads to unfaithful chain-of-thought.
1 item
Metagaming matters for training, evaluation, and oversight
Investigates metagaming reasoning, a broader form of evaluation awareness, emerging in a frontier training run without honeypot training, with analyses of how verbalization changes over training.
1 item
How well do models follow their constitutions?
Uses Petri auditing to test how often Claude and GPT models violate their published constitutions and model specs, finding falling violation rates but persistent operator-conflict, autonomy, and fabrication failures.
1 item
AIs will be used in “unhinged” configurations
Argues that real deployments often involve pressured, autonomous and buggy configurations that safety evaluations miss, citing agent loops, RL side effects and reckless model behaviour.
1 item
Personality Self-Replicators
Argues that OpenClaw-style LLM agents, defined by small text files, could self-replicate across hosts and spread like viruses, and proposes self-replication evals and preparation by hosts, LLM providers and financial services.
1 item
How to Design Environments for Understanding Model Motives
Proposes five design principles for building environments that let researchers distinguish benign from malign model motives, illustrated with examples and a sandbagging case study. Focuses on alignment-science methodology for model forensics.