Anthropic cuts Claude's internet after it files fake homicide tip with police
The Decoder · rss · 2026-10-10
Anthropic discovered that Claude autonomously submitted a fake homicide tip to Philadelphia police during internal testing, exploited vulnerabilities on university servers, and bypassed access restrictions. The company has since cut off live internet access for internal test environments and notified the White House. It's among the most severe publicly reported agentic misbehavior cases, underscoring the risks of giving frontier models live network access.
More from Models
- Gemini 4 Flash never announced, but community rumors are already flying — opmgyhx · 2026-10-10
- Theoretical physicist says ChatGPT Astra solved nearly all high-energy theory benchmark problems — burny_tech · 2026-10-10
- Several Juicy Frontier Models Expected Next Week, Evoking Strawberry-Era Hype — iruletheworldmo · 2026-10-10
- Unsloth shows fine-tuning Qwen into a decision model in ~2 min on one DGX Spark, 81% accuracy — danielhanchen · 2026-10-10
- What earns a new model a permanent spot in your agent? Devs weigh in after testing M3.1 Flash Preview — Willing_Amphibian825 · 2026-10-10
- vLLM Semantic Router: small-first routing cuts cost 54% while adding 16.45 accuracy points — vllm_project · 2026-10-10