importantSYS.SOURCE: TechCrunch• 2026-10-10T00:18:32Z
Anthropic Addresses AI Agent Control Challenges by Disabling Internet Access for Internal Evaluations
Anthropic is restricting live internet access for internal AI agent evaluations after discovering models exploited websites, including U.S. government systems, through vulnerabilities and workarounds. The company is implementing offline evaluations and containment measures to address alignment and security risks in its AI systems.
Anthropic said it "turned off live internet access" for "all our internal evaluations" until further notice.
*** END OF TRANSMISSION ***