Key facts
- Hackers are stealing Claude AI tokens by exploiting compromised session keys.
- Infostealer malware is used to steal user login sessions and access Claude accounts.
- Anthropic identified unauthorized third-party activity on affected accounts.
- Affected users experienced unexpected token consumption and unauthorized account upgrades.
- Anthropic suspended accounts, invalidated sessions, and issued partial refunds to some users.
Hackers are stealing Claude AI tokens by exploiting compromised session keys, impacting user accounts and business operations. AI consultant Grant de Swardt discovered unauthorized use of his Claude Max account, leading to unexpected token consumption and unauthorized upgrades. Anthropic confirmed that hackers use infostealer malware to steal user login sessions and access Claude accounts, though the malware does not originate from Claude itself.
Anthropic has suspended affected accounts, invalidated sessions, and issued partial refunds. De Swardt's experience was not isolated, with other users reporting similar issues. The company stated that the malware can be acquired from various online sources. De Swardt has since switched to Cursor and expressed concerns about the lack of transparency in usage tracking and tools for users to protect themselves.
Separately, Anthropic reported a fourth cybersecurity incident in January involving an early version of its Claude Opus 4.6 model, notifying affected parties. This follows previous disclosures of AI models being hacked during cybersecurity tests and broader scrutiny of AI companies over AI breakout events.