We reviewed the “Risks from automated R&D” section of Anthropic’s February 2026 Risk Report, producing two corresponding review documents: our original review and our updated review.…
Summary: We describe three different definitions of the productivity impact of AI (AKA uplift), and show there’s reason to expect: [text{uplift on old tasks} leq text{uplift…
.post-content p:has(> img:only-child) { margin-top: 0 !important; margin-bottom: 0 !important; } .post-content p:has(> img:only-child) + p, .post-content h3 + p:has(> img:only-child) { margin-top: 0 !important; }…
This is a linkpost for MirrorCode, a project that METR funded and co-developed with Epoch AI. See Epoch AI’s blog post for more detail: https://epoch.ai/publications/mirrorcode-preliminary-results/
p:has(> img) { margin-bottom: 0; } .content img { margin: 0.75em 0; } Kei Nishimura-Gasparian is an Astra fellow and was the primary contributor to this…
Update: Further details on this exercise are included in our Frontier Risk Report (February-March 2026), within the Anthropic section of Appendix B. In collaboration with Anthropic,…
As METR’s time horizon task suite saturates, the results are becoming more sensitive to analysis choices. One example of this was the recent update to fix…
Introduction METR aims to keep the public informed about the capabilities of and risks posed by AI — by some metrics the fastest-moving technology in history,…
We reviewed two versions of Anthropic’s Sabotage Risk Report for Claude Opus 4.6, producing two corresponding review documents: our review of the February 11 version and…
.content figure figcaption p { font-weight: normal; } Summary: We find that roughly half of test-passing SWE-bench Verified PRs written by mid-2024 to mid/late-2025 agents would…