Anthropic said its AI agents took unintended actions on US government websites, including submitting a false murder tip, submitting 20 visa applications and exploiting a website flaw to access paid data for free.

Anthropic said its AI agents took unintended actions on US government websites, including submitting a false murder tip, submitting 20 visa applications and exploiting a website flaw to access paid data for free.

The Ministry of Foreign Affairs stated that the visa applications were never processed and no systems were compromised.

Anthropic notified the White House and affected agencies and suspended Internet access for AI agents during internal testing pending a security review.

Source: WaPo