METR
10 analyses mentioning METR, newest first.
METR's Updated Time Horizon Plot Shows AI Still Lags on Judgment-Heavy Tasks
METR's updated time horizon report finds exponential gains on verifiable tasks, while frontier models still trail humans on open-ended, judgment-based work like designing evaluations.
Anthropic Goes Public on Nasdaq, Citing Safety Concerns Amid IPO
Anthropic listed on Nasdaq, revealing significant losses but emphasizing a governance structure that prioritizes AI safety and human control over shareholder demands.
OpenAI Embeds Apollo Researchers to Test AI Subversion Risks
OpenAI is piloting the embedding of Apollo safety evaluators to monitor for model subversion attempts, following recent safety incidents and industry calls for deeper oversight.
OpenAI Launches Deep, Independent 'Embedded' Safety Assessments with Key Partners
OpenAI has partnered with four independent assessors to conduct deep, embedded evaluations of its internal models and safeguards, aiming to boost public transparency regarding AI risk management.
New Congressman Pushes Hard for Federal AI Regulation Amid Existential Risk Fears
Newly elected Congressman Scott Wiener is aggressively advocating for strong federal AI regulation, citing concerns over recursive self-improvement and the need for legislative action before potential AI catastrophe.
AI Coding Boom Masks Skill Atrophy: The 'Supervision Paradox' in Software Engineering
New research suggests over-reliance on AI coding tools can stunt junior developers' core skills, creating a 'supervision paradox' where proficiency declines despite perceived productivity gains.
Anthropic Grants Watchdog 'Employee-Like' Access to Assess Model Risks
Anthropic has allowed the nonprofit METR to embed a team within its offices with deep access to its hardware and systems to independently audit model safety, signaling a move toward external accountability.
Anthropic Grants Deep Access to METR Researchers for Independent AI Audits
Anthropic is granting six researchers from METR, a nonprofit, employee-level access to independently audit its models and safety practices, signaling a commitment to external accountability.
Anthropic CEO Amodei Proposes Global Pause and Regulatory 'Embedded' Oversight for AI Development.
Anthropic CEO Dario Amodei advocates for a global slowdown in AI development, committing Anthropic to a new model of third-party, regulator-like embedded evaluation.
AI Coding Hackathon Reveals Unexpected Findings: AI Slows Developers, Human Creativity Still Wins
A recent San Francisco hackathon pitting AI-supported teams against humans revealed surprising results: AI tools actually hampered experienced developers' productivity, while human teams demonstrated creative problem-solving and nuanced understanding—despite the AI-supported team winning the top prize.

