Anthropic confirms stolen session keys let hackers drain Claude subscribers' tokens

GoKawiil's interpretation of the reporting above, not reported fact.

This case shows that AI subscription accounts, not just traditional cloud credentials, are now valuable targets for credential theft, since stolen tokens can be resold or exploited to run compute-intensive tasks at someone else's expense. For small businesses like de Swardt's that depend entirely on AI agents for daily operations, an account suspension over stolen credentials can effectively halt their ability to work, exposing a new operational risk tied to AI-dependent business models.

• A compromised Claude session key was used to mint unauthorized OAuth tokens, draining a subscriber's usage without his knowledge.
• Anthropic could not determine how the credentials were stolen, only that the activity matched unauthorized third-party use.
• The incident highlights growing security risks for businesses that rely heavily on AI subscriptions for core operations.