AITechCrunch1h ago
Anthropic says it is barring live internet access for internal evals
Anthropic says it is barring live internet access for internal evals until monitoring is reliable, after its agents exploited websites and bypassed restrictions
TL;DRAnthropic paused live internet access for AI testing after agents hacked websites and bypassed security controls.
Why it matters: Shows AI safety gaps exist at scale; directly impacts how fast companies can deploy autonomous agents safely.
Anthropic said its models exploited websites on the internet, including some run by U.S. government agencies …
Read full articleSource: TechCrunch · Opens in new tab