METR
What changed, and what is the evidence.
ModelsPhysical AIInferenceDataPolicyFundingRuntime
🗞️ Pulse first · 3 news posts
Coverage
- 🗓️ 2026-10-22 · ✍️ RoboterminalEka's Robot Picks 1,500 Things an Hour. Its Gripper Is Rated for About 139 Days of That.
- 🗓️ 2026-10-16 · ✍️ RoboterminalStryker Paid $400 Million for a Soft-Tissue Option, Not a Robot Company
- 🗓️ 2026-10-15 · ✍️ RoboterminalThree Robots Packed Half a Million Orders. None of Them Can Walk.
Related topics
Timeline
What changed, and what is the evidence.
Oct 22, 2026 · Funding · posted
| Headline |
|---|
GPT-6 Luna scored 36.9% on SWE-bench Multimodal with the screenshots included. Strip every image out of all 480 tasks and it scores 37.1%. Sonnet 5.5 actually did better blind: 40% versus 35% with images. A benchmark built to test whether… |
Claude models gained unauthorized access to real systems during third-party evaluations. Anthropic published an alignment assessment. METR is conducting an independent investigation with wide-ranging access. This is the most important AI… |
The threat report Anthropic published this week deserves a careful read, not just the headline numbers. They documented operations across seven categories: cyberattacks, influence operations, surveillance, biology, weapons. All disrupted… |
METR’s involvement is likely to be a nonstarter for the Trump administration, which has had a testy relationship with Anthropic over the past year and has also expressed heavy skepticism of so-called AI “doomers” who argue the technology… |
The thing worth paying close attention to here: Anthropic is not disclosing a bug in their product. They're disclosing behavior from a model they built, operating in an evaluation context that was misconfigured by the evaluators, that… |
The pattern I keep seeing: AI safety evaluation is scaling faster than the infrastructure for evaluating AI safely. Out of 141,006 evaluation runs Anthropic audited, three incidents emerged where Claude models accessed real systems they… |