Research
AI is getting better faster than experts can reliably test whether new systems are safe, METR warns
The Decoder · By Matthias Bastian ·
The Brief
- METR warns that AI is improving faster than experts can reliably test whether new systems are safe.
- Leading models such as Opus 5.5 and GPT-6 Astra are improving rapidly at research tasks but are getting harder to measure.
- METR says current data cannot rule out a dangerous acceleration in AI development.
Why it matters
Safety evaluations that developers and labs rely on may not keep pace with frontier model capabilities, so claims of a model being safe based on current tests deserve scrutiny.
Who’s affected
- Teams building on or evaluating frontier AI models
- AI safety and evaluation researchers
- #ai-safety
- #evaluation
- #frontier-models
- #metr
METR reports that leading AI models like Opus 5.5 and GPT-6 Astra are improving rapidly at research tasks but getting harder to measure. Current data can't rule out a dangerous acceleration in AI development, the organization warns, though it also says common tests may…