Anthropic is issuing a security warning to some Claude AI users after it emerged that malware, which is responsible for stealing authenticated browser session data, is being used to compromise these accounts. When attacked by this kind of malware, Anthropic say that adversaries are then able to sign into the victim’s Claude account using the stolen session data, enabling them to use Claude and burn through your usage without needing to use your password.
The event brings to light growing cyber security concerns surrounding AI services where simply compromising the active web sessions belonging to a victim on their internet browser can grant you access to their accounts without the need for them to even input their passwords.
Anthropic warns Claude users about malware attack
According to Anthropic, the threat actor at hand is using mainstream infostealer malware in order to exfiltrate live Claude sessions, steal it from compromised computer systems and subsequently utilize the stolen, authenticated browser sessions as a mechanism for accessing associated Claude accounts. As stated in the company’s latest announcement of an Anthropic security update concerning malicious activity observed while interacting with Claude AI, these bad actors can then potentially use these browser sessions as proof of authentication in order to use user accounts without necessarily knowing how to unlock them.
Anthropic has reportedly notified users on compromised systems of suspicious account activity as well. Users may even note that usage has been unexpectedly reset and then quickly depleting back toward zero, even if said user hasn’t even been logging into Claude, or if you simply forgot to log out of an active browsing window previously authenticated with Claude.
In response, Anthropic is automatically signing effected users out of Claude and removing saved payment details, while also offering refunds for unauthorized charges that may be assessed.
How does infostealer malware target Claude accounts?
Infostealer malware are malware programs that aim to covertly harvest sensitive information from your personal computer, but browser cookies and login session data are a favorite target. This can include just about any form of personally sensitive information found on your browser. The Anthropic user account security concerns described above are a special kind of malware targeting an online service with a sophisticated and constantly improving web experience. In particular, when attackers go about hacking your Claude sessions with it, they can sign in from your account, consume its usage, without knowing what your password for the account actually is.
Because the current malware campaign targeting Claude sessions doesn’t require attacker knowledge of your user password, no matter how rigorous your password complexity standards are, you are vulnerable if your machine has been compromised. The malware used in this campaign is thought to be of either the Vidar, LummaC2, StealC, RedLine, Acreed malware families targeting users running the Windows operating system or potentially Atomic Stealer (AMOS) for users on Mac systems.
Why Claude users should take the warning seriously
For the most part, whether you’re using the advanced AI system for coding, researching, to make yourself a writing aid for content creation purposes, or even for professional tasks like legal or medical assistance, security and financial aspects are certainly not negligible. An unauthorized individual who gains possession of your account could exhaust its full usage. And if your payment details are tied to the account, you could end up paying more than anticipated. The experience also teaches you something more fundamental than a phishing alert: how protecting only the password isn’t always adequate; in fact, modern threats actively exploit this security gap.
What should affected Claude users do?
There’s one point especially important from Anthropic’s warning: signing out of just Claude from your personal computer does absolutely nothing to get rid of the malware you may currently have running, once again, if that malware resides upon your personal computer or Mac system. Here’s a breakdown of the advice given by Anthropic: Users should always attempt to scan their computer and delete any existing and active malware present on the system after becoming aware of such an issue; then proceed to the remaining user directives below.
Other related security actions for affected Claude users:
Change your email account’s password linked to Claude.
Log out of all of your active browser sessions from across all of your devices; consider ending your authenticated browser sessions from your personal computers too, although signing them out of Chrome, Edge, Firefox, or Safari alone won’t delete your Malware.
Turn on two-factor authentication whenever possible.
Reset all of your current and old passwords for other useful sites like social media or banking services that may have sensitive data associated with your profile.
Always review bank statements and/or credit card statements for fraudulent transactions after being notified of something like this.
Review and check to be safe against saved passwords and billing methods in any of your browsers, make sure not to leave payment methods logged in.
It’s crucial never to utilize and/or download unauthorized software or use unofficial website links found over the web, just like illegal downloads.
After you have successfully managed to make safe your personal computer or any computing device associated, you’ll need to add your payment details again to use Claude.
Users must never re-authorize Claude from compromised machines; otherwise the same outcome can occur again and potentially from those exact compromised machines.
Is Claude itself responsible for the malware?
Anthropic emphasizes strongly there is no evidence of malware stemming from the Claude platform itself but the compromise appears to occur because of a virus existing already on the users own computers that is actively searching for credentials belonging to various online platforms. From there the AI session is one more cookie to extract from the logged in user so they have no idea how to compromise anything beyond simply being signed in using the stolen cookie data. This isn’t an incident that points to something malicious being supplied to the user. Instead, this is malware that’s already there and can steal session information for various online services all at once.
Security fears regarding AI are growing rapidly
This security alert appears at an especially sensitive time, as cyber security researchers and several AI businesses have found increasing concerns over secure operating processes of advanced AI models, particularly since earlier in this month Anthropic admitted that their Claude models’ use cases had inadvertently provided direct system access after misconfiguration of certain security elements and inadvertently granting some testing models unintended internet access, leading to multiple breaches of real systems. Anthropic says they will perform additional checks on more than 141,000 testing runs since that point and confirmed three confirmed instances where user systems had undergone a breach during these instances. While their first example involves a data loss type situation, the incident described in the security warning today still shows how users of their system can still find their accounts exploited despite efforts taken on the side of Anthropic.
Bottom Line
Users and supporters of the Claude AI will learn some important security measures on how they need to make sure their usage limits are never used for malicious purposes by hackers by logging in to your Claude account using the password associated with it. But beyond that, infostealer malware can quietly and surreptitiously compromise your devices, leading directly to user session theft from browsing applications. Users who spot abnormal activity within their Claude account or observe abnormal use of their Claude usage will need to protect their system, sign out entirely, and proceed to resetting these credentials. Protecting the device used is as crucial as securing the application itself going forward.
