Multiple Anthropic Claude Max subscribers reported discovering that their monthly usage allowances were being consumed without any action on their part — one user's account moved from 0 per cent to 49 per cent used in twelve minutes. Anthropic investigated and identified infostealer malware as the attack vector: malicious software spread via infected software downloads and malicious advertising harvests saved Claude session credentials from infected machines and creates unauthorised OAuth tokens, which attackers then use to consume the victim's subscription quota. Anthropic has suspended affected accounts, invalidated compromised sessions, and issued partial refunds; the company declined to provide itemised usage logs or announce new protective measures. For practitioners running agent workflows or client projects through Claude subscription tiers, session credential theft is now a distinct threat category from API key compromise — the attack surface is the browser's saved credential store, not the application's secrets management. Recommended immediate steps: enable two-factor authentication on your Anthropic account, audit active OAuth applications in account settings, and set up anomaly alerts on usage dashboards. Unusual token consumption that cannot be explained by your own workflows should be treated as a potential compromise indicator, not an attribution error.