Hackers Target Anthropic Claude Users to Steal Tokens

9 Sept 2026, 2:40 am · 4d ago · 1 min read · TechCrunch AI

Anthropic has issued a warning to subscribers after malicious actors began compromising user accounts to steal Claude tokens. The issue came to light after a subscriber discovered their account was consuming tokens and running automated tasks while inactive. Attackers exploit compromised user credentials or session tokens to route heavy generative AI queries without paying, burning through subscribers' paid quotas. Security researchers emphasize the growing vulnerability of AI model credentials as API tokens become high-value targets for digital thieves seeking compute power. Anthropic is advising users to update credentials and enable enhanced security protocols.