Security · 1h ago
Researchers show Anthropic's Claude Cowork can escape sandbox, access Mac files
Security researchers demonstrated that Anthropic's Claude Cowork agent can break out of its restricted environment and access arbitrary files on a Mac. The attack exploits the model's ability to execute shell commands and read system files. Anthropic has partially mitigated the issue, but users can take additional steps to protect themselves.
Meridian48 take
This highlights a fundamental challenge with AI agents: giving them tool access inevitably creates new attack surfaces that are hard to fully lock down.
Read the full reporting
It's not just OpenAI models escaping and running riot — experts show how Claude Cowork can break its bonds and access Mac files →
TechRadar
ai-safetyanthropic