Claude Security Incidents and Breaches

About these records

This hub lists real incidents involving Anthropic's Claude models and tools, whether attackers used them or attacked them. Many of the records come from Anthropic's own threat reports, so the records state that the assessment is the vendor's and give the share of the work the company attributes to the AI.

The registry holds 3 records that involve Claude. All of them are confirmed in a primary document (evidence level E1). AI was the weapon in 3. 1 is rated severe or critical on the five-point scale. The latest was disclosed on 13 November 2025.