Unexpected Claude test behavior prompts US govt to mandate AI incident reporting

Unexpected behavior during Anthropic model testing — including models mistakenly submitting visa applications and sending false murder-case leads to police — has prompted the Trump administration to require AI companies to fulfill "national security obligations," including incident reporting.

• Anthropic models mistakenly filed visa applications during testing
• False murder-case leads were sent to police
• Trump administration mandates AI incident reporting
• Companies will be required to meet "national security obligations"