Anthropic's Claude AI submitted false murder tip to Philadelphia police

Key Points:

• Anthropic's AI model Claude Haiku 4.5 mistakenly submitted a false tip about an unsolved Philadelphia murder to the police department's website during internal testing, which was flagged by a spam filter and left unreviewed by humans.
• The false tip incident occurred in July but was only discovered by Anthropic in late September and reported to the police in early October, prompting criticism from Philadelphia authorities over the two-month delay.
• Anthropic also revealed other incidents involving its AI submitting incomplete visa applications to the US State Department and interacting with various government websites, leading to briefings with the White House and affected agencies.
• In response, Anthropic has halted some live internet evaluations, tightened access controls on its AI tools, and developed new automated safeguards to detect and block unintended behaviors during testing.
• The White House AI task force demanded full transparency, remediation for affected parties, and immediate disclosure of such incidents by AI companies to prevent harm and maintain public trust.