Tracker ID CR-R04
On loss of control: "Current systems lack the capabilities to pose such risks, but they are improving in relevant areas." Models increasingly distinguish test from deployment.
Reference
- Claim made
- Who
- International AI Safety Report 2026
- Target date
- No target date set
- Status checked
- 28 September 2026
Source
Evidence to date
Assessment. Related trend: METR time-horizon doubling about every 7 months (2019–2025), about 89 days since 2024 (METR TH1.1, 29 Jan 2026).
Next indicator
AI Safety Report 2027
Where this claim appears
- Fears vs Reality: Will we lose control of AI? · Article
- Fears vs Reality: Could AI threaten humanity itself? · Article
- If AI becomes far more capable than us, how would we live alongside it? A calm guide to a speculative question · Article
- AI Camps: The sceptics · Article
- AI Camps: The safety-focused · Article
- AI Camps: Risks, and the conditions they would need · Article
- AI Camps: The accelerationists · Article
- AI Camps: Abundance, and what it would take · Article
Found this useful? Save it to your basket.