We are Changing our Developer Productivity Experiment Design
Our second developer productivity study faces selection effects from wider AI adoption, prompting us to redesign our approach.
Five lessons from having helped run an AI-Biology RCT
Luca Righetti shares takeaways on the role of randomized controlled trials in AI safety testing.
How We Protect Confidential Information
Our high-level approach to protecting confidential access and information
Analyzing coding agent transcripts to upper bound productivity gains from AI agents
Amy Deng investigates whether coding agent transcripts could serve as an alternative for estimating AI productivity uplift, using 5305 Claude Code transcripts from METR technical staff.
Measuring Time Horizon using Claude Code and Codex
Nikola Jurkovic describes our measurements of time horizon using Claude Code and Codex scaffolds.
A simpler AI timelines model predicts 99% AI R&D automation in ~2032
Thomas Kwa describes a simple model for forecasting when AI will automate AI development, based on the AI Futures model but with only 8 parameters.
Frontier AI safety regulations: A reference for lab staff
Miles Kodama and Michael Chen summarize key provisions from California's SB 53, the EU Code of Practice, and New York's RAISE Act covering frontier AI developers.
Time Horizon 1.1
We’re releasing a new version of our time horizon estimates (TH1.1), using more tasks and a new eval infrastructure.
Early work on monitorability evaluations
We show preliminary results on a prototype evaluation that tests monitors' ability to catch AI agents doing side tasks, and AI agents' ability to bypass this monitoring.
Clarifying limitations of time horizon
Thomas Kwa responds to some misinterpretations of our time horizon work, and explains limitations and the core finding.
Common Elements of Frontier AI Safety Policies (December 2025 Update)
Shared components of AI lab commitments to evaluate and mitigate severe risks.
Details about METR's evaluation of OpenAI GPT-5.1-Codex-Max
We evaluate whether GPT-5.1-Codex-Max poses significant catastrophic risks via AI self-improvement, rogue replication, or sabotage of AI labs. We conclude that this seems unlikely.