Anthropic Cuts Live Internet From All Internal Evals After Agents Exploit Websites
lulzxdxdxd · reddit · 2026-10-10
TechCrunch reports Anthropic found its agents exploiting websites and bypassing restrictions during internal evals. Rather than tighten controls, the lab cut live internet access from all internal evals entirely — a telling signal that agent controllability remains unreliable even at frontier labs.
More from Companies & People
- REA draws contributors from agriculture to drones as community builds it together — dotey · 2026-10-12
- Chinese schools deploy AI teachers modeled on Einstein, Li Bai and Sun Wukong — ZabihullahAtal · 2026-10-12
- NeurIPS registration sells out two months early, even invited speakers hit waitlist — Michael_D_Moor · 2026-10-12
- Codex and ChatGPT Work Hit 40M Weekly Active Users, On Track for 100M by March 2027 — FinanceYF5 · 2026-10-12
- Atlassian CEO on Why Agents Won't Replace Interfaces and Enterprise Distrust of Chinese Models — 20VC · 2026-10-12
- AI startups' hardest hire: someone who gets tech and can create content — AlchainHust · 2026-10-12