Anthropic's August 2025 misuse report describes an extortion operation in which Claude helped turn stolen information into targeted demands. The company says an attacker targeting at least 17 organizations used Claude Code during intrusions, then used Claude to assess stolen financial data and choose ransom amounts. This made the AI part of the operation's decision-making process. Anthropic says it banned the accounts and added detection measures. The published case summary supports that account of misuse; its displayed ransom examples are simulated illustrations made by Anthropic's team, not original attacker messages.
Terms in this piece · Glossary
agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
