Listen to this Post
A New Kind of AI Account Theft
Anthropic is warning some Claude users that a dangerous class of malware known as infostealers has been stealing active Claude login sessions directly from infected computers. The attackers can then reuse those authenticated sessions to access Claude accounts, consume AI usage, and potentially make unauthorized purchases if payment information is stored on the account.
The warning highlights a growing problem in the AI era: stealing a password is no longer the only way to take over an account. If malware can capture an already authenticated browser session, an attacker may be able to bypass the normal login process—including the point where a user would normally be protected by two-factor authentication.
Anthropic says it has identified several well-known infostealer families associated with compromised users, including Vidar, LummaC2, StealC, RedLine, Acreed, and Atomic Stealer (AMOS) on a small number of Mac systems.
The company is now taking direct action against affected accounts by revoking stolen sessions, removing saved payment methods, and refunding charges it determines were unauthorized.
Anthropic’s Warning to Claude Users
Anthropic reportedly contacted affected Claude customers after detecting suspicious activity associated with their accounts. The company explained that attackers were using stolen Claude sessions obtained from infected computers rather than exploiting a vulnerability in Claude itself.
One of the most important clues is unusual usage behavior.
If a Claude user notices that their usage allowance appears to refill and then rapidly disappears while they are not actively using the service, Anthropic says this could indicate that somebody else is operating the account.
That makes unexplained usage spikes more than just an annoying billing or quota problem. In some circumstances, they could be evidence that an attacker has obtained an authenticated session.
The Malware May Have Nothing to Do With Claude
Anthropic has emphasized an important distinction: there is currently no indication that these infections originated from Claude itself.
The malware appears to be general-purpose information-stealing software that was already present on affected Windows or Mac computers.
Infostealers are designed to collect as much valuable information as possible from a compromised machine. Depending on the malware, that can include browser passwords, authentication cookies, cryptocurrency wallets, autofill information, credentials, tokens, and data belonging to other applications.
Claude sessions can simply become another valuable item in that collection.
How a Stolen Claude Session Can Become an Account Takeover
Traditional account theft often starts with a stolen username and password.
Infostealers can take a different route.
When a user logs into a web service, the browser normally maintains authentication information that tells the website that the user has already successfully authenticated.
An attacker who obtains the relevant session information may be able to reuse it from another environment.
This is why simply changing a password after an infection may not always be enough.
If an attacker already possesses a valid session token, the password change might not immediately invalidate that token unless the service explicitly revokes active sessions.
That is precisely why Anthropic is revoking compromised Claude sessions.
Why Two-Factor Authentication Is Not a Complete Shield
Two-factor authentication remains one of the most important protections available to users.
But there is a crucial limitation.
2FA generally protects the authentication event. If an attacker steals an already authenticated session after the user has successfully completed that authentication, the attacker may not need to repeat the password-and-2FA process.
This does not make 2FA useless. Quite the opposite—it remains extremely valuable against password theft and phishing.
But it demonstrates why modern account security increasingly requires more than simply protecting passwords.
Session protection, device security, token revocation, suspicious-login detection, hardware-backed authentication, and endpoint protection all matter.
Vidar, LummaC2, StealC and RedLine Enter the Picture
Anthropic says its investigation has identified multiple infostealer families associated with affected systems.
Among them are Vidar, LummaC2, StealC, and RedLine, alongside Acreed on Windows systems.
A small number of Mac systems were also associated with Atomic Stealer, commonly known as AMOS.
These malware families have different technical characteristics, but their fundamental objective is similar: compromise a device and extract information that can be monetized or abused.
The Claude sessions apparently became one more valuable credential type for criminals to harvest.
The Pirated Software Problem
One affected Claude user who shared
That does not prove that every affected user became infected through pirated software.
However, it illustrates one of the oldest cybersecurity lessons: software obtained from untrusted sources can become an extremely dangerous entry point.
Pirated games, cracked applications, unofficial installers, fake updates, malicious browser extensions, and modified utilities are all commonly abused because users expect them to execute with significant privileges on their computers.
The attacker does not necessarily need to exploit Claude when the user’s own computer can provide access to the account.
Signing Out Is Only the First Step
Anthropic’s response demonstrates an important security principle: revoking the stolen session is not the same thing as cleaning the infected device.
Signing the user out of Claude can invalidate the particular session that an attacker has stolen.
But if the infostealer remains installed, the next time the user logs back into Claude, the malware could potentially steal the newly created authentication data again.
This creates a dangerous cycle:
Infection → login → session theft → account takeover → logout → login again → session theft again.
The only reliable way to break that cycle is to address the underlying malware infection.
Anthropic Is Removing Saved Payment Methods
The threat is not limited to Claude usage.
If a compromised account has saved payment information, attackers may attempt to exploit that access for unauthorized purchases or paid usage.
Anthropic says it is removing saved payment methods from affected accounts as a precaution.
The company also says it will refund charges that its investigation determines were unauthorized.
That response is significant because it recognizes that account compromise can quickly become a financial-security issue rather than merely a privacy problem.
What Claude Users Should Do Immediately
Anyone who receives a similar warning should treat it as a device compromise, not simply an account-password problem.
First, stop using the potentially infected computer for sensitive logins.
Second, remove the malware using a trusted security product or, when appropriate, completely reinstall the operating system.
Third, change important passwords from a known-clean device.
Fourth, revoke active sessions wherever the service provides that capability.
Fifth, review payment activity and account usage for anything unexpected.
Finally, do not immediately return to the infected machine and log into every account again.
Otherwise, the attacker may simply collect the replacement credentials.
Deep Analysis: How Infostealers Can Steal AI Sessions
Inspect Active Sessions
Users should review active sessions for important accounts whenever the service provides this option.
For example, after cleaning a machine, check browser sessions and account-security dashboards for unfamiliar devices, locations, or authentication events.
For Claude specifically, users should follow
Check Running Processes on Windows
Windows users investigating suspicious behavior can inspect running processes with:
Get-Process | Sort-Object CPU -Descending | Select-Object -First 25
This does not prove that a process is malicious, because legitimate applications can consume significant CPU.
It is simply a starting point for identifying unusual processes.
Review Recently Installed Applications
PowerShell can help identify recently installed software:
Get-ItemProperty <code>HKLM:\Software\Microsoft\Windows\CurrentVersion\Uninstall\,</code> HKLM:\Software\Wow6432Node\Microsoft\Windows\CurrentVersion\Uninstall\ | Select-Object DisplayName, DisplayVersion, Publisher, InstallDate |
Sort-Object InstallDate -Descending
Unexpected applications, especially software installed shortly before suspicious account activity, deserve investigation.
Check Network Connections
Administrators can inspect active TCP connections with:
Get-NetTCPConnection | Where-Object State -eq "Established" | Select-Object LocalAddress,LocalPort,RemoteAddress,RemotePort,OwningProcess
You can then investigate a suspicious process ID:
Get-Process -Id <PID>
Network connections alone cannot identify malware, because legitimate applications constantly communicate with remote servers.
They can, however, provide useful evidence during an investigation.
Linux Users Can Inspect Network Activity
On Linux systems, administrators can use:
ss -tupn
To examine processes and network connections together, another useful command is:
sudo lsof -i -P -n
Again, these commands are diagnostic tools—not malware scanners.
Look for Persistence
Infostealers frequently attempt to survive reboots.
Windows users investigating a potentially compromised machine can review startup applications:
Get-CimInstance Win32_StartupCommand | Select-Object Name, Command, Location, User
Suspicious persistence mechanisms should be investigated carefully rather than deleted blindly.
Browser Security Matters Too
Because infostealers frequently target browsers, browser security should be treated as part of the overall account-security strategy.
After a confirmed infection, assume that browser-stored passwords, cookies, autofill information, and active sessions may have been exposed.
Do not assume that clearing browser history alone solves the problem.
A compromised machine can potentially capture credentials again after the browser is cleaned.
The Safest Response to a Confirmed Infection
For high-value systems, a complete operating-system reinstall may be preferable to attempting to manually remove sophisticated malware.
The general sequence should be:
1. Isolate the infected computer
2. Stop sensitive logins
3. Preserve evidence if investigation is required
4. Clean or rebuild the device
5. Change credentials from a trusted device
6. Revoke active sessions
7. Re-enable strong MFA
8. Review account activity
9. Monitor for renewed compromise
The critical concept is that credential recovery should happen after device recovery whenever possible.
Why AI Accounts Are Becoming Attractive Targets
AI Usage Has Economic Value
Cybercriminals have historically targeted email accounts, banking credentials, social-media profiles, cloud dashboards, and cryptocurrency wallets.
AI accounts are increasingly joining that list.
A stolen AI account can provide access to paid usage, organizational resources, proprietary conversations, API-related information, and potentially sensitive business material.
The value of an AI account therefore depends on much more than its subscription price.
Attackers Can Abuse Legitimate AI Infrastructure
A compromised AI account can potentially be used to perform large amounts of legitimate-looking activity.
That creates an interesting security problem.
From the
The request may arrive with valid session information.
The attacker may not need to exploit an obvious vulnerability.
Instead, the criminal simply becomes the user.
AI Makes Stolen Sessions More Interesting
There is another emerging dimension.
AI systems can process information at enormous scale.
An attacker who compromises a business account may attempt to use AI tools to analyze stolen information, generate content, automate repetitive tasks, or consume large quantities of computing resources.
This means account security increasingly protects not only personal conversations but potentially an entire productivity environment.
What Undercode Say:
The Password Is No Longer the Whole Story
The most important lesson from this incident is that cybersecurity has moved beyond password protection.
A password can be strong.
2FA can be enabled.
A user can still become compromised if malware takes control of an already authenticated endpoint.
Session Theft Is Quietly Dangerous
Session theft is particularly concerning because it can look legitimate.
The attacker does not necessarily need to trigger a failed login.
They may inherit an authenticated state.
That makes traditional login alerts less useful in some scenarios.
Endpoint Security Has Become Account Security
The Claude incident demonstrates how closely endpoint security and cloud security are now connected.
A secure cloud service cannot fully protect an account if the endpoint holding its authentication material is compromised.
The browser has effectively become part of the security perimeter.
Infostealers Are Built for Scale
Infostealers are attractive to criminals because they can collect large amounts of information automatically.
One infection may expose dozens of accounts.
Email.
Social media.
Cloud services.
Shopping accounts.
Cryptocurrency wallets.
AI platforms.
Corporate credentials.
Everything potentially becomes valuable.
The AI Industry Is Entering a New Security Phase
Claude is not unique in facing this problem.
As AI services become central to work and personal productivity, criminals have increasing incentives to target AI identities.
The more valuable the account becomes, the more attractive session theft becomes.
Usage Spikes Can Become Security Signals
Anthropic’s warning about suddenly depleted usage limits is particularly interesting.
Unusual consumption can potentially function as an early warning signal.
AI providers could increasingly use behavioral analytics to identify impossible or highly unusual usage patterns.
Behavioral Detection Could Become Essential
Future AI security systems may analyze:
Typical usage times
Device fingerprints
Geographic patterns
Request frequency
Token consumption
Session behavior
API patterns
Model selection
Conversation characteristics
The objective would be to distinguish the legitimate user from someone who merely possesses their session.
But Detection Has to Be Careful
Aggressive security systems can also create false positives.
A legitimate user might suddenly work late at night.
They might travel.
They might upgrade their subscription.
They might have an unusually productive day.
Security systems therefore need multiple signals rather than relying on a single anomaly.
Revocation Is Becoming More Important
Services need fast and reliable session revocation.
If a token is stolen, users should be able to terminate it quickly.
Ideally, organizations should also have centralized controls that allow administrators to invalidate sessions across entire environments.
MFA Needs Stronger Foundations
MFA remains essential.
But phishing-resistant authentication methods and hardware-backed credentials can provide stronger protection than traditional password-and-code combinations.
The broader goal should be to make authentication credentials difficult to steal and difficult to replay.
Browser Security Deserves More Attention
Users often think of the browser as merely an application.
Modern browsers are actually enormous repositories of identity information.
They contain sessions, cookies, credentials, tokens, extensions, autofill data, and access to dozens of cloud services.
That makes browsers extremely attractive targets.
Pirated Software Remains a Major Risk
The affected Reddit
Attackers do not always need sophisticated zero-days.
Sometimes they simply need a user to install something that looks useful.
Convenience Creates Attack Surface
Every saved password, remembered login, browser extension, and automatically authenticated service increases convenience.
But convenience also creates additional opportunities for malware to steal access.
Modern security is therefore increasingly about balancing usability with containment.
AI Accounts Should Be Treated Like Critical Accounts
Users should stop thinking about AI subscriptions as disposable entertainment accounts.
For many people, AI services now contain work documents, private conversations, research, business information, and intellectual property.
That makes them valuable targets.
Companies Need Endpoint and Cloud Controls Together
Organizations should combine endpoint detection with identity monitoring.
Blocking malware is important.
Detecting abnormal cloud behavior is equally important.
The two systems become significantly stronger when they share information.
Account Takeover Can Become a Supply-Chain Problem
One compromised employee account can potentially expose information belonging to an entire organization.
That means an individual endpoint infection can become an organizational incident.
Security teams should therefore treat AI credentials as corporate assets where appropriate.
AI Providers Have a New Responsibility
AI companies need to assume that some legitimate sessions will eventually be stolen.
That means designing systems around the assumption that authentication tokens can be compromised.
Short-lived sessions, strong token binding, device verification, rapid revocation, and behavioral monitoring can reduce the damage.
Users Also Have a Responsibility
No security architecture can compensate for downloading arbitrary executable software and granting it access to a personal computer.
Users remain a critical part of the security chain.
The safest approach is simple: install software from trustworthy sources and treat unexpected installers with suspicion.
A Clean Device Is the Foundation
Changing every password while leaving an infostealer installed is like changing the locks while the burglar is still inside the house.
The device must be trusted again before credentials can be considered safe.
Session Revocation Should Become Routine
After a suspected compromise, users should revoke active sessions wherever possible.
This is especially important for email, cloud storage, password managers, AI platforms, developer tools, and financial services.
AI Usage Theft Could Become a Larger Criminal Market
As AI models become more powerful and expensive, compromised accounts could become valuable commodities.
Criminals may eventually specialize in stealing AI subscriptions just as they currently specialize in cryptocurrency wallets, cloud accounts, and social-media identities.
Security Teams Should Monitor Consumption
For enterprise AI deployments, sudden increases in token usage should be investigated.
Unexpected model usage can indicate compromised credentials, automation abuse, insider misuse, or a misconfigured application.
The AI Identity Perimeter Is Expanding
The modern identity perimeter now stretches from the user’s hardware to the browser, cloud session, AI provider, API credentials, and connected applications.
Every component matters.
A weakness in one layer can undermine the others.
The Biggest Lesson Is Simple
The Anthropic warning is not really a story about Claude being hacked.
It is a story about what happens when an endpoint becomes compromised and the attacker discovers valuable authenticated sessions inside it.
That distinction matters.
Infostealers Turn One Infection Into Many Breaches
One infected computer can expose multiple online identities simultaneously.
The damage therefore scales far beyond the original malware infection.
A single malicious download can become an account-takeover event across an entire digital life.
AI Security Will Need to Evolve
The AI industry is moving toward increasingly sophisticated identity and abuse detection.
The future will likely involve continuous authentication rather than authentication only at login.
In other words, systems will increasingly ask:
Does this activity still look like the legitimate user?
The Era of “Login Once and Forget” Is Fading
Persistent sessions are convenient.
They are also attractive targets.
As AI services become more valuable, users and providers may increasingly need to balance long-lived sessions against stronger security controls.
Security Is Becoming Continuous
The most important takeaway is that security does not end when a user successfully logs in.
It continues throughout the entire session.
It continues while the browser is open.
It continues while the device is connected.
And it continues after suspicious activity appears.
✅ Anthropic Has Warned About Stolen Claude Sessions
The supplied report accurately describes Anthropic warning affected users that infostealer malware had been used to steal active Claude sessions and consume account usage.
The response is consistent with the incident details provided in the original article.
✅ Infostealers Can Steal Browser Authentication Data
Modern infostealers commonly target browser-stored credentials, cookies, session information, and other locally accessible secrets.
That makes an infected endpoint a serious risk even when the victim uses strong passwords and MFA.
✅ Signing Out Does Not Remove Malware
This is an important and technically sound distinction.
Revoking a stolen session can prevent reuse of that particular session, but it does not disinfect the computer that originally leaked it.
✅ Multiple Infostealer Families Were Identified
The original report identifies Vidar, LummaC2, StealC, RedLine, Acreed, and Atomic Stealer among the malware associated with affected systems.
That does not mean every affected user had all of these malware families; they represent malware observed across the investigated cases.
❌ A Claude Session Theft Incident Does Not Mean Claude Was Hacked
The available information does not establish that attackers exploited a vulnerability in Claude itself.
Anthropic specifically indicated that the affected computers were likely infected with general-purpose infostealers.
The attack appears to have targeted authentication material residing on compromised endpoints.
Prediction
(+1) AI Providers Will Build Stronger Continuous Session Protection
AI companies are likely to invest heavily in behavioral detection, rapid session revocation, device-bound authentication, suspicious-usage monitoring, and stronger identity controls.
As AI accounts become more valuable, authentication will increasingly evolve from a one-time login event into a continuously monitored security process.
The most successful platforms will not simply ask whether a user has the correct credentials—they will increasingly evaluate whether the behavior behind those credentials makes sense.
(+1) Infostealers Will Target AI Services More Aggressively
Claude is unlikely to be the last AI platform targeted by credential-stealing malware.
As users store more valuable information and consume increasingly expensive AI resources, stolen AI sessions could become an increasingly attractive commodity for cybercriminals.
The lesson for users is straightforward: protect the device, protect the browser, protect the session, and only then protect the password.
▶️ Related Video (82% Match):
🕵️📝Let’s dive deep and fact‑check.
🎓 Live Courses & Certifications:
Join Undercode Academy for Verified Certifications
🚀 Request a Custom Project:
Secure, high-velocity infrastructure and disruptive technological engineering. Contact our engineering team for high-tier development and proprietary systems:
[email protected]
💎 Smart Architecture | 🛡️ Secure by Design | ⭐ Trusted by Thousands
References:
Reported By: www.bleepingcomputer.com
Extra Source Hub (Possible Sources for article):
https://www.digitaltrends.com
Wikipedia
OpenAi & Undercode AI
Image Source:
Unsplash
Undercode AI DI v2
🔐JOIN OUR CYBER WORLD [ CVE News • HackMonitor • UndercodeNews ]
📢 Follow UndercodeNews & Stay Tuned:
𝕏 formerly Twitter 🐦 | @ Threads | 🔗 Linkedin | 🦋BlueSky | 🐘Mastodon | 📺Youtube
