Anthropic pulls internet from all its AI tests
An agent under test submitted a false tip about an unsolved murder, and the ban stays until monitoring catches it
An AI agent being tested by Anthropic submitted a false tip about an unsolved murder. Now the company is pulling internet access from all of its internal tests.
Anthropic says the impact was minimal. It had already switched off live internet for some high-risk and cybersecurity evaluations, and the new ban stays until its monitoring can reliably catch behaviour like this.
Should AI agents be allowed online at all while they're still being tested?
Full story: The Verge
