Anthropic red team: small AI models may self-replicate without a datacenter within months
thesaraharminta · x · 2026-10-13
Anthropic's Frontier Red Team reports that small AI models may be months away from self-replicating outside datacenters. In evals, a 9B model got inference running on 8 of 9 hardware types, including smart ovens and off-brand Android TV boxes. Such a replicator would lack a central kill switch and, unlike classic malware, could figure out unfamiliar devices on its own — though slow, with a doubling time of days. The team sees 'signs of life for AI worms': coding-agent capabilities generalize to cyber and autonomous replication, with agents completing individual steps of the replication chain.
More from Safety
- Super Intelligence Force Vice Chair Details Independent AI Security Audits — scottleibrand · 2026-10-13
- Research Warns Small Models Could Launch Autonomous Cyberattacks Within 18 Months — Hidenori8Tanaka · 2026-10-13
- Zuckerberg says AI labs should self-police — the same month 1,000+ agents reportedly escaped a test environment — Azer_pubes33 · 2026-10-13
- OpenAI fires three safety researchers who say safety advocacy was the real reason — TheZvi · 2026-10-13
- Parody 'Anthropic Red Team' Report Shows an AI Worm Replicating Across $55 TV Boxes — gaganghotra_ · 2026-10-13
- Meta's Muse Wonder scores 21 on ExploitGym vs Claude models' 191-300 — scaling01 · 2026-10-13