Anthropic Refunds Users After 5 Infostealers Drain Claude Usage

AI platform operators and enterprise users have spent heavily on model safety and prompt injection defenses, but this incident shows a more mundane attack vector can consume Claude usage and drain budgets. The risk is not a sophisticated AI exploit; it is commodity infostealer malware harvesting an authenticated browser session and silently using your account.

On August 30, 2026, Anthropic began publicly confirming that infostealer malware is being used to hijack active Claude login sessions, letting attackers access accounts and consume usage. The company is signing affected users out of Claude, removing saved payment methods, and refunding charges it identifies as unauthorized. While this surfaced as a customer notice, it is a meaningful security event because the compromise does not exploit a vulnerability in Claude's model or infrastructure. Instead, it repurposes a long-standing cybercrime technique against a high-value AI service account.

“

Anthropic identified multiple Windows infostealer families connected to the activity: Vidar, LummaC2, StealC, RedLine, and Acreed.

The mechanism is familiar to defenders but still dangerous. Infostealers are malicious programs that silently steal passwords, browser cookies, login credentials, and other sensitive data from an infected computer, then transmit that data to criminals. Unlike ransomware, which announces its presence by encrypting files and demanding payment, infostealers are stealthy by design, allowing enough time to collect a broad set of credentials. In this case, attackers have specifically begun pulling Claude sessions from larger infostealer log collections and reusing them. Because an infostealer can copy an already authenticated browser session, the attacker may not need to pass the normal password and 2FA login process. That means even users with strong authentication could still lose access to their Claude account and have usage drained.

Anthropic's email, sent to affected users and shared on Reddit, says the company has "recently become aware of a bad actor that is using common infostealer malware to steal Claude login sessions from people's computers, then using those login sessions to access Claude accounts and consume their usage." Users whose usage limits looked like they refilled and then drained while they weren't using Claude were likely seeing this activity. One Redditor who posted the notice said they had downloaded a pirated game, which likely explains how their system was compromised. Anthropic stressed that it has "no reason to believe that this malware is related to Claude, installed through Claude, or related to anything you did with Claude." The malware typically arrives through downloads or malicious apps and steals information stored locally, including browser passwords, login cookies, and credentials belonging to other apps.

Anthropic identified multiple Windows infostealer families connected to the activity: Vidar, LummaC2, StealC, RedLine, and Acreed. Naming these families matters because they are all established infostealer-as-a-service tools, widely used by lower-skilled actors who buy or rent access. The attackers are now monetizing stolen AI sessions, which is a notable evolution from simply selling credential logs on dark web marketplaces. By hijacking a Claude session, an attacker can consume the victim's paid usage, drain prepaid credits, or potentially access API capabilities tied to the account. For organizations that use Claude through individual browser sessions, the financial and operational impact could be immediate and difficult to distinguish from legitimate usage without monitoring.