Safety & RiskPartially correct
“A straightforward extrapolation of today's systems to those we expect to see in two to three years suggests a substantial risk that AI systems will be able to fill in all the missing pieces [of bioweapons production], enabling many more actors to carry out large-scale biological attacks.”
Co-founder and CEO, Anthropic
Said Jul 25, 2023Deadline Jul 25, 2026Written testimony of Dario Amodei, U.S. Senate Judiciary Subcommittee on Privacy, Technology, and the Law, 'Oversight of A.I.: Principles for Regulation', July 25, 2023
Directionally borne out on capability trend but not on the strongest version of the claim: by May 2025 Anthropic activated ASL-3 safeguards for Claude Opus 4 because it could not rule out CBRN-3 (meaningful uplift to novices in bioweapons acquisition planning, measured at ~2.5x), and OpenAI has treated recent agentic models as 'High' biological capability. However, as of the mid-2026 deadline no frontier model has been publicly assessed as able to fill in 'all the missing pieces' (state-level/CBRN-4 thresholds remain uncrossed), and no AI-enabled large-scale biological attack has occurred — partly consistent with Amodei's own conditional that guardrails and mitigations be put in place.