- Anthropic invalidated active Claude user sessions and wiped saved payment cards after infostealers compromised account tokens.
- Six malware families across Windows and macOS extracted browser cookies, allowing attackers to bypass multi-factor authentication and drain usage limits.
- Affected users must fully clean local devices and secure primary email accounts before logging back in or adding new payment methods.
Artificial intelligence startup Anthropic has initiated emergency account protection measures for a group of Claude users after detecting widespread session hijacking. Cybercriminals used active login sessions stolen directly from user devices to gain unauthorized access and drain account usage limits.
Internal monitoring systems identified the suspicious account activity and triggered immediate defensive protocols across affected profiles. Company notifications confirmed that infostealer malware families targeting both Windows and macOS operating systems enabled the unauthorized access.
How Malware Steals Session Cookies to Hijack Accounts
This security incident is due to infostealer malware found in infected computers rather than the hacking of the servers of Anthropic. This malicious software easily collects the saved passwords, cookies, and active authentication tokens from valid login sessions. By replaying these stolen session cookies, attackers bypass standard password prompts and two-factor authentication controls entirely.
Subsequently, threat actors used the hijacked sessions to interact with Claude models and burn through paid account allowances. Impacted customers first noticed the intrusion when their daily usage limits refilled and depleted rapidly at hours when they were not active.
Notifications sent to affected account holders identified six distinct stealer malware families responsible for collecting local credentials. The Windows-based threats included Vidar, LummaC2, StealC, RedLine, and Acreed, while Atomic Stealer targeted macOS systems.
Emergency Response and Account Mitigation Measures
To halt ongoing unauthorized usage, Anthropic automatically invalidated all active login sessions associated with compromised accounts. Shutting down these connections right away means that people stealing cookies are now out of luck. This holds as long as the perpetrators did not break into the servers that store cookies.
The firm has also assured that it has removed any saved credit details in those accounts. Thus, it minimizes risks for unauthorized use of such information, be it in future transactions or automatic renewals.
In addition to revoking access, Anthropic has also refunded the amounts paid without the consent of the customers. According to the instructions from the company, affected users should restore their payment information only after they complete hardware cleaning.
Despite the efficiency and effectiveness of server-side logouts, experts point out that such a solution prevents attacks on active connections but does not eliminate malware from the computers of customers. Users who log back in on an infected computer risk having their fresh session tokens stolen again immediately.
Operational Vulnerabilities in Token-Based Web Authentication
The principle of the token-based web authentication system is that many modern online platforms depend on cookies to keep users logged. This occurs without prompting them to input their passwords every time they log on to the system.
When a subscriber logs into an artificial intelligence platform, server infrastructure generates a unique digital token stored within the local web browser. Every subsequent query or request sends this token back to platform servers to verify user identity.
However, local browser security models cannot prevent malicious software executing with user-level privileges from reading cookie stores. Infostealer malware routinely scans browser directories, copies session tokens, and transmits the data back to central command servers.
Since these tokens represent pre-authenticated state records, attackers can import them into secondary browsers to impersonate victims instantly. Consequently, traditional security perimeters fail to detect session replay attacks because incoming queries carry valid authentication credentials.
The stolen session tokens and credentials that fuel attacks like the Claude hijacking campaign are part of a much larger criminal ecosystem. AI-generated phishing campaigns now achieve click rates of 54% compared to just 12% for traditional campaigns, and stolen credentials have become the second most common initial access vector for intrusions as infostealers flood criminal markets with data from compromised machines.
The Rising Threat of Session Hijacking in the AI Sector
The targeting of Claude accounts highlights a broader trend where cybercriminals monetize stolen session access to high-value cloud services. As paid artificial intelligence subscriptions become standard tools for developers and businesses, access tokens represent valuable digital assets. Infostealer operators frequently bundle collected session cookies into automated marketplaces for fast resale to secondary threat actors.
Furthermore, session hijacking presents special problems for traditional exterior defenses due to the fact that the traffic may be perceived as originating from a legitimate authenticated user. Traditional multi-factor authentication works only during the authentication phase and makes the post-authentication cookies vulnerable to endpoint malware.
Thus, it is essential for artificial intelligence vendors to implement advanced behavioral analytics to detect unusual behavior originating from unknown IP locations or device configurations.
Essential Remediation Steps for Affected Users
The IT security teams provide clear protocols on remediation that victims should follow before they can have their accounts reinstated. The first step is to run full antivirus checks on all local devices to detect and isolate the existing info-stealing agents. Disconnecting compromised systems from local networks will limit the ability of malware to send additional batches of data to remote servers during remediation.
Next, individuals should secure the primary email account associated with their Claude profile from a verified clean device. Changing email passwords and generating hardware-based two-factor authentication should stop the attackers from gaining access to the account through the password recovery links.
Finally, the users are free to log in to the platform again. Also, if necessary, they can create new developer API keys and provide their billing information once again. The users should implement strong password management and refrain from downloading any unverified software to avoid infections in the future.
