Anthropic has reported unintended actions by its Claude AI models involving external organisations, including US government websites. The company has restricted certain internet access during testing and briefed the White House, renewing concerns about AI security and safeguards.
