METR

For three weeks, from July 21st to August 11th, a METR employee and a Redwood Research employee conducted a pilot high-access exercise to stress-test OpenAI’s internal misalignment monitoring system. 1 The output of this exercise was a 61-page internal report that was shared with

Executive summary

This article discusses how an assessor could study whether capabilities progress could accelerate dramatically due to AI-driven acceleration of AI R&D. Such an assessment would answer: Capabilities Progress : How fast are internal capabilities improving, and how much has this pro

Recent AI misalignment incidents have shown AI systems capably pursuing goals their human supervisors would not approve of, like hacking other companies. Fortunately, current AIs still seem relatively bad at concealing their misbehavior from human reviewers: these recent misalign

On September 30, 2026, METR President Chris Painter testified before the U.S. Senate Committee on Homeland Security & Governmental Affairs’ Subcommittee on Disaster Management, District of Columbia, and Census, at a hearing titled “Rogue AI: Securing the Homeland Against AI Agent

In light of recent incidents (e.g. those from OpenAI , Anthropic , and UK AISI ), we developed and deployed a basic live per-action monitor to reduce the likelihood of incidents involving harmful actions from agents during our own evaluations. The monitor is intended to reliably

Note on independence: This evaluation was conducted under an unpaid agreement for AI R&D assessment. 1 We drafted the initial summary, and then Anthropic had the opportunity to review and edit the text. We signed off on this final text from the Claude Opus 5.5 system card . Our p

Please note that this post focuses on incidents where external actors attempted to gain unauthorized access to METR’s systems, not AI agents hacking in our evaluations. We have conducted an initial scan of our evaluations, and currently have no evidence of any agents hacking thir

Core takeaways about this incident

Q: How much has discovery accelerated? LLMs have shown the ability to make novel discoveries across many domains. How much has this affected the aggregate discovery rate? In the figures below we plot all the data sources we can find, and make some very loose observations: Discove

Update (September 5, 2026) : We’ve updated our suggested questions for investigators to make them more precise and cover limitations. See the update log for the original questions. AI agents sometimes autonomously take sophisticated, sustained actions in clear violation of user a

**Goals of this post** > > **Goal:** a simple way to compare a variety of capability metrics in an idealized setting where we observe only a function from $x$ to $s$. > > Non-goals: > > 1. Treatment of practical difficulties (an incomplete list of other considerations appears at

METR disclosed two external security incidents earlier in 2026 and described investigative findings and security-process changes. It reported that incident #1 involved stolen API credentials leading to substantial illicit usage worth approximately $600,000 in credits (though those credits were granted to METR for free).

METR published a post describing an independent effort (with Redwood Research staff contracting with METR) to understand model behavior observed in a recent OpenAI/Hugging Face incident, outlining the scope and high-level terms of the engagement.

METR reported commitments of around $71 million raised over the prior six months, describing high-level intended areas of work enabled by the funding.

METR published proposed questions and framing for thorough external investigations after misalignment incidents, including an update log and context for intended investigation outputs.

METR summarized its independent, predeployment evaluation of GPT-5.6 Sol, including how evaluation outcomes depended on detection and handling of cheating attempts.

METR released a frontier risk report describing a pilot exercise to assess misalignment risks from AI agents used inside frontier AI developers, including a stated participation model with Anthropic, Google, Meta, and OpenAI and a framework for the assessment process.