AI R&D is accelerating. The control evidence is not keeping pace
New reporting points to faster model development, but the strongest claims still rest on uncertain and partly redacted evidence.
2026-11-09 · 5 min · 5 sources
METR
x · Follow ↗
Evals & red teaming
Frontier-model evaluations and autonomous-task capability
- Type
- Organization account
- Priority
- Tier 1 · core
- Perspective
- Independent evaluations
- Best for
- Evaluation results and methodology updates
- Publishes
- Frequent
- Editor's note
- High-value early signal; read full methodology.
Our analysis featuring METR
Embedded AI evaluators need access, publication rights, and visible limits
Anthropic's new arrangement and METR's assessment proposal give buyers a way to ask what an outside review actually covers.
2026-10-19 · 5 min · 3 sources
The real AI acceleration signal is organizational, not benchmark-based
Three recent accounts point to a better executive test: measure how AI changes the R&D system, not just how models score.
2026-09-28 · 6 min · 9 sources
Recent from METR
- https://t.co/ozhLbSzud0 [LINKED EVIDENCE: https://metr.org/blog/2026-10-06-ai-systems-could-cover-u 2026-10-06
- Treating agent observability as security-critical infrastructure means handling all AI outputs (tran 2026-10-06
- This bug did not change the underlying logs, just what the viewer showed. More capable agents could 2026-10-06
- This demonstration used a JavaScript injection via MathJax rendering, which an agent could trigger f 2026-10-06
- Here’s a simple example: a METR researcher found a bug in the transcript viewer of Inspect, a popula 2026-10-06
- In order to notice when AI agents misbehave, AI companies often log the actions and reasoning steps 2026-10-06