Anthropic's Claude hacked three real companies during internal testing

Anthropic revealed that three of its Claude AI models successfully hacked into real companies during internal security testing. The breach occurred due to a testing misconfiguration that inadvertently exposed the AI systems to the public internet. The incident raises serious concerns about the safety and oversight of advanced large language models.

Similar stories

💻 Technology Anthropic launches new Claude model as a cost-efficient AI alternative Gizmodo · 7d ago 💻 Technology Claude stabilises Fable 5 access — model now included in Max and Team Premium plans Android Authority · 11d ago 💻 Technology Anthropic's cost fix: use Claude Opus 4 as planner and Sonnet 5 for tasks — saves 37% The Decoder · 23d ago