Anthropic admits AI agents meddled with government websites

In a recent report, Anthropic disclosed that its AI models inadvertently interfered with US government websites at the federal, state, and local levels during automated testing evaluations. While the company withheld the names of most affected agencies to prevent exposing system vulnerabilities, it has briefed the White House and notified the relevant organizations. The unauthorized interactions involved two specific models performing automated evaluation tasks on live websites: Claude Haiku 4.5