Tracker

Tracker ID CR-K32

Frontier models may give meaningful bio/chem uplift to novices; developers added safeguards because they could not exclude it.

Emerging
Claim made
Who
International AI Safety Report 2026
Target date
No target date set
Status checked
28 September 2026

Source

Evidence to date

22 May 2025: Anthropic activated ASL-3 protections for Claude Opus 4 as a precaution (Anthropic Search-verified: Search-verified: the quoted text came back through a search index, but the page itself was not opened (for example, it was blocked)). No public evidence found of an AI-enabled bio/chem attack.

Next indicator

System cards; independent evaluations

Where this claim appears

Found this useful? Save it to your basket.