METR
METR estimates Anthropic’s AI R&D progress has nearly doubled
Anthropic is likely improving its models on well-specified tasks at roughly twice the rate recorded between March 2024 and October 2025, according to an independent…
METR Says Anthropic May Have Crossed an AI R&D Warning Threshold
METR says Anthropic’s models posed a low risk of directly automating key research and development or causing acute catastrophic risks as of July 15, 2026. However, the…
METR and Redwood stress-test OpenAI's internal agent monitoring, find gaps
A METR employee and a Redwood Research employee spent three weeks probing OpenAI's internal misalignment monitoring system and found a number of novel vulnerabilities…
METR agrees Anthropic's misalignment risk is low, but doubts the methods
METR, an external reviewer, agrees with Anthropic that the risk of catastrophic harm from misalignment is low for the models in Anthropic's August 2026 Risk Report. It…
METR Outlines How an Outside Assessor Could Probe Hidden AI Misalignment
METR has published a proposal for an "alignment assessment" in which an outside assessor would try to understand the worst misaligned actions a developer's AI agents…
METR sets out rules for how it works with AI companies on risk evidence
METR, a nonprofit that gathers evidence on whether humans could lose control of AI agents at scale, has published a post describing how it works with AI companies on…
METR Outlines How Assessors Could Test for AI-Driven Acceleration in AI R&D
METR has published a proposal for how an outside assessor could investigate whether AI automation of AI research and development is causing capabilities progress to…
METR Urges AI Developers to Open Internal Systems to Loss-of-Control Assessments
METR, a group that evaluates frontier AI, has published a proposal asking AI developers to let independent assessors examine whether they could lose control of their…
METR Warns AI Agents Could Hide Misbehavior by Tampering With Review Tools
AI safety group METR says the tools used to review what AI agents do should be treated as security-critical infrastructure, because a capable agent could tamper with…