Anthropic has forced sign-outs for Claude users whose accounts were compromised by infostealer malware, according to the company’s notifications to affected customers. The AI firm said the malware harvested active Claude login sessions from users’ own computers, not from a breach at Anthropic. Attackers who gained access could drain usage limits and make unauthorized charges on the accounts. Anthropic’s email to affected users said that if their usage limits appeared to refill and then deplete while they were not using Claude, that was likely the cause.
In response, Anthropic terminated the affected sessions, removed the saved payment card on file, and refunded any extra usage charges tied to the malicious activity. The company warned that this action does not remove the malware from the user’s device, and logging back in without cleaning the machine could lead to another compromise. Anthropic advised users to remove the malware first, then update the password on the linked email account and enable two-factor authentication. Only after those steps should users re-add their payment methods to Claude.
The incident highlights a growing trade in stolen AI account credentials. Adam Meyers, senior vice president of counter adversary operations at CrowdStrike, told Axios in August that criminals have developed a market for hijacked accounts for AI services such as Claude, ChatGPT, and Gemini. Palo Alto Networks’ Unit 42 research team traced hijacked accounts to proxy services called transfer stations, which pool stolen credentials and resell access to AI tools at prices well below normal retail rates. These findings underscore that the value of an AI login extends beyond the account itself, because the account can be used to generate content or consume computing resources at the victim’s expense.
Anthropic does not publish exact figures for Claude usage limits, but each prompt carries a compute cost that the company absorbs when a stolen session runs queries. An attacker with a hijacked session and a saved payment method can also purchase additional usage on the victim’s account. That explains why Anthropic deleted stored payment methods rather than simply terminating the stolen sessions. The company earlier doubled rate limits for paid Claude Code users in May because demand had been outpacing capacity, a sign of how expensive and contested AI compute resources have become.
TechManNews reached out to Anthropic for comment on the sign-outs and the malware incident but did not receive a response. The company’s guidance to affected users focuses on securing the endpoint before returning to the service, a step that applies to any cloud platform with stored credentials. U.S. users of Claude who suspect their accounts were hijacked should check for unusual usage patterns and follow the remediation steps outlined by Anthropic. The broader lesson for American consumers is that malware on a personal device can silently capture active sessions for remote services, bypassing password protections entirely.
More cybersecurity news from TechManNews.
