Claude Accounts Targeted By Infostealer Malware, Anthropic Warns

Anthropic has warned some Claude users that infostealer malware has stolen active login sessions from their computers, allowing attackers to access their Claude accounts and consume paid usage quotas. The company said it detected unauthorized activity involving compromised Claude sessions and has responded by signing affected users out, revoking the stolen sessions and removing saved payment methods. Anthropic also said it will refund charges that its investigation determines were unauthorized.

The incident does not appear to involve a breach of Anthropic’s own systems. Instead, the company says the attackers obtained Claude browser sessions from already-infected computers using general-purpose infostealer malware. These malicious programs can steal browser cookies, saved passwords and other credentials, allowing attackers to reuse an authenticated session without necessarily entering the victim’s password or completing two-factor authentication again.

Anthropic Warns Claude Users About Stolen Sessions

Anthropic has begun notifying users whose Claude accounts showed signs of unauthorized access.

The company said a “bad actor” had been using common infostealer malware to steal active Claude login sessions and then use those sessions to consume the affected accounts’ available usage.

One warning sign is particularly notable: users may see their Claude usage limits refill and then rapidly disappear even when they were not actively using the service.

Claude Session Theft At A Glance

Issue	Reported Details
Target	Claude user accounts
Attack method	Stolen browser sessions
Initial compromise	Infostealer malware
Main impact	Unauthorized Claude usage
Payment risk	Saved payment methods potentially exposed to unauthorized charges
Anthropic response	Session revocation and sign-out
Payment response	Saved payment methods removed
Refunds	Unauthorized charges to be refunded where identified
Anthropic infrastructure breach	No evidence reported
Investigation	Ongoing

Anthropic has not disclosed how many Claude accounts have been affected.

How Infostealer Malware Hijacks Claude Accounts

Infostealer malware is designed to quietly collect information from an infected computer.

Depending on the malware, stolen information can include browser passwords, cookies, authentication tokens and credentials stored by applications.

In this case, Anthropic says the malware collected Claude’s active login session along with other information on compromised machines.

Attack Chain

Malicious download
       │
       ▼
Infostealer infects computer
       │
       ▼
Browser cookies + credentials stolen
       │
       ▼
Claude session copied
       │
       ▼
Attacker reuses authenticated session
       │
       ▼
Claude account accessed
       │
       ▼
Usage quota consumed

The key point is that the attacker does not necessarily need to steal the user’s Claude password directly.

Stolen Sessions Can Bypass Password And 2FA

A browser session represents an already authenticated state.

Once a user logs into a website, the browser stores information that allows the service to recognize that user during subsequent requests.

If malware copies the relevant session information, an attacker may be able to reuse it.

That means the attacker can potentially avoid the normal login process.

Normal Login Vs Session Hijacking

Normal Access	Stolen Session
Enter credentials	Reuse stolen session
Password checked	Existing authentication reused
2FA may be requested	May not need to repeat 2FA
New session created	Victim’s session effectively copied
User controls login	Attacker can impersonate active session

This is why session-token theft remains a serious security problem even when users have strong passwords and two-factor authentication enabled.

Anthropic Says The Malware Was Not From Claude

Anthropic emphasized that the malware involved in the incident was not related to Claude itself.

The company said it had no reason to believe the malware was installed through Claude or that the incident resulted from something users did within the Claude service.

Instead, Anthropic described the malware as general-purpose infostealer software that typically arrives through unofficial downloads or malicious applications.

Where Infostealers Can Come From

Unofficial software
       │
       ├── Pirated applications
       ├── Cracked games
       ├── Fake utilities
       ├── Malicious installers
       └── Other untrusted downloads
              │
              ▼
        Infostealer infection

The exact infection method can vary between victims.

One affected user who shared Anthropic’s notification reportedly said they had downloaded a pirated game before discovering that their computer had been compromised.

Malware Families Identified By Anthropic

Anthropic said its investigation had identified several known infostealer families on affected systems.

The malware included Vidar, LummaC2, StealC, RedLine and Acreed on Windows systems.

The company also identified Atomic Stealer, or AMOS, on a small number of Mac computers.

Malware Identified

Platform	Infostealers Reported
Windows	Vidar
Windows	Lumma / LummaC2
Windows	StealC
Windows	RedLine
Windows	Acreed
macOS	Atomic Stealer (AMOS), in a small number of cases

These are established malware families associated with credential and information theft.

Their presence on a computer does not by itself prove that Claude was the original target.

Why AI Accounts Are Becoming Valuable Targets

Claude accounts can contain significant amounts of paid or subscription-based usage.

For attackers, compromising such an account can create an opportunity to consume someone else’s AI quota.

In some cases, attackers may also attempt to exploit saved payment methods or use compromised accounts for other purposes.

Potential Value Of A Compromised AI Account

Asset	Potential Attacker Interest
AI usage quota	Generate unauthorized requests
Paid subscription	Consume included capacity
Saved payment method	Attempt unauthorized purchases
Account identity	Abuse the account
Browser session	Maintain access without credentials
Connected services	Potential secondary exposure

Anthropic’s response specifically addresses the first two risks by revoking sessions and removing stored payment information.

Anthropic Is Signing Affected Users Out

Anthropic said it has been revoking compromised sessions and signing affected users out.

This invalidates the stolen session, meaning the attacker should no longer be able to use that copied session to access Claude.

The company is also removing saved payment methods from affected accounts.

Anthropic’s Response

Detect suspicious activity
          │
          ▼
Identify affected account
          │
          ▼
Revoke stolen sessions
          │
          ▼
Sign user out
          │
          ▼
Remove saved payment method
          │
          ▼
Investigate unauthorized charges
          │
          ▼
Refund eligible charges

Anthropic said affected users may need to log back into Claude on their devices after the security action.

Signing Out Does Not Remove The Malware

This is one of the most important warnings in Anthropic’s notification.

Revoking a Claude session stops the stolen session, but it does not remove the malware that originally stole it.

If the infostealer remains active on the computer, a newly created Claude session could potentially be stolen again after the user logs back in.

Why Malware Removal Is Essential

Compromised computer
        │
        ▼
Existing Claude session revoked
        │
        ▼
User logs in again
        │
        ▼
Infostealer still active
        │
        ▼
New session potentially stolen
        │
        ▼
Attack repeats

This makes device cleanup just as important as changing account credentials.

Anthropic Recommends Security Cleanup

Affected users are being urged to take additional security measures.

These include removing the malware from their computers, changing relevant credentials and revoking sessions.

Users should also review their accounts for suspicious activity.

Recommended Actions

Action	Purpose
Remove malware	Stop further credential theft
Change passwords	Protect compromised credentials
Revoke active sessions	End unauthorized access
Check account activity	Identify suspicious usage
Review payment activity	Detect unauthorized charges
Avoid untrusted downloads	Reduce reinfection risk
Reinstall affected system if necessary	Ensure persistent malware is removed

The exact remediation required depends on the severity and type of infection.

Claude Users Can Check Active Sessions

Anthropic’s current account-security tools allow users to view and manage active Claude sessions.

The Claude Help Center says users can navigate to Settings > Account > Active sessions to see devices and browsers currently logged into their account. Each session includes information such as the device, browser, approximate location and last update.

Users can terminate an unfamiliar session directly from that interface.

Claude Session Management

Claude
  │
  ▼
Profile
  │
  ▼
Settings
  │
  ▼
Account
  │
  ▼
Active sessions
  │
  ├── Device
  ├── Browser
  ├── Approximate location
  └── Last activity
          │
          ▼
Terminate suspicious session

Anthropic also provides an option to log out of all active web sessions through the account settings.

Claude Web Sessions Can Last For 28 Days

Anthropic’s Help Center says Claude web sessions have a 28-day duration when inactive, with the session automatically refreshed when users take actions such as reloading Claude.

This makes session security particularly important because a stolen active session can remain useful for an extended period if it is not revoked.

Claude Session Details

Session Attribute	Current Information
Web session duration	28 days
Activity effect	Can refresh session
Session management	Available
Remote termination	Available
Logout all sessions	Available on web
Mobile logout-all feature	Not currently available through mobile apps

The existence of session-management controls gives users a way to identify and terminate suspicious access.

Phones And Tablets Do Not Appear To Be Involved

Anthropic’s notification reportedly said its investigation had so far identified compromised computers, while phones and tablets did not appear to be involved in the campaign.

This does not mean mobile devices are immune to malware generally.

Rather, it reflects Anthropic’s findings in this particular incident.

Reported Device Exposure

Device	Reported Involvement
Windows PCs	Yes
Macs	Small number identified
Phones	Not appearing involved
Tablets	Not appearing involved

The company continues to investigate the incident.

Why Infostealers Are Particularly Dangerous

Traditional account attacks often attempt to obtain usernames and passwords.

Infostealers can collect much more information at once.

A compromised browser may contain authentication cookies, saved passwords, autofill information and other data that can provide attackers with access to multiple services.

Information Infostealers Can Target

Infected computer
      │
      ├── Browser cookies
      ├── Saved passwords
      ├── Authentication tokens
      ├── Autofill information
      ├── App credentials
      └── Other local data

This means a Claude compromise could be only one symptom of a much larger device-security problem.

Users Should Check Other Accounts Too

If an infostealer is discovered on a computer, users should not assume that only their Claude account is affected.

Because these malware families can collect credentials and cookies from multiple applications, other online services may also be at risk.

Potentially affected accounts could include email, cloud storage, social networks, developer platforms and financial services.

Broader Account-Security Checklist

Account Type	What To Review
Email	Recent logins and sessions
Cloud storage	Active devices
Developer accounts	API keys and tokens
Social media	Login history
Financial services	Transactions
AI platforms	Usage and active sessions
Password manager	Security alerts and credentials

If a device is confirmed to be infected, simply changing one password may not be sufficient.

AI Usage Theft Creates A New Cybersecurity Problem

The incident highlights an emerging category of cybercrime: AI resource theft.

Attackers do not necessarily need to steal money directly.

They can exploit another person’s paid AI subscription or API capacity.

The resources can then potentially be used for activities that consume substantial computing capacity.

AI Account Abuse

Stolen AI session
       │
       ▼
Unauthorized account access
       │
       ▼
AI usage consumed
       │
       ├── Coding
       ├── Research
       ├── Content generation
       └── Other workloads
       │
       ▼
Victim's quota depleted

This creates a new financial incentive for attackers to target AI accounts.

Claude Is Not The Only AI Service At Risk

The underlying attack technique is not specific to Anthropic.

Any online service that relies on persistent browser sessions can potentially be exposed if an attacker successfully steals the relevant authentication data from an infected machine.

That means the lesson extends beyond Claude.

Session Hijacking Risk

Security Layer	Possible Attack
Password	Stolen directly
2FA	Bypassed through stolen active session
Browser session	Copied by malware
API credential	Extracted from local system
Saved payment data	Potentially exposed
Connected applications	Potential secondary risk

Strong authentication remains important, but it cannot fully protect a device that has already been compromised by malware capable of stealing authenticated sessions.

The Incident Highlights Endpoint Security

Anthropic’s warning ultimately points to a broader cybersecurity issue.

Cloud services can maintain strong server-side security while users’ own devices remain vulnerable.

If an attacker controls the endpoint, they may be able to access information that is already authenticated.

Secure cloud service
        │
        ▼
Secure account
        │
        ▼
Compromised user computer
        │
        ▼
Authenticated session stolen
        │
        ▼
Unauthorized account access

This is why endpoint security remains a critical part of modern account protection.

Anthropic’s Security Response Is Still Developing

Anthropic said its investigation is ongoing.

The company has identified several malware families and has implemented account-level protections for users it believes were affected.

However, the full scope of the incident has not been disclosed.

There is currently no public figure for the number of compromised accounts.

Known Vs Unknown

Question	Current Status
Were Claude sessions stolen?	Yes, Anthropic says so
Was infostealer malware involved?	Yes
Was Anthropic’s infrastructure breached?	No evidence reported
Malware families identified?	Yes
Number of affected users	Not disclosed
Investigation complete?	No
Unauthorized charges refunded?	Eligible charges identified by investigation
Are affected devices safe after logout?	Not necessarily

The distinction between account remediation and device remediation is particularly important.

The Bigger Picture

The Claude session theft warning demonstrates how cybersecurity risks are evolving alongside the rapid adoption of AI services. Attackers reportedly used common infostealer malware to steal authenticated browser sessions from compromised computers and then consume victims’ Claude usage. Anthropic says the incident was not caused by a vulnerability in Claude or by malware distributed through Claude; instead, the stolen sessions appear to have been collected from already-infected devices.

The incident also shows why session security has become increasingly important for AI platforms. A stolen authenticated session can potentially allow an attacker to bypass the normal password and two-factor authentication flow, while the victim may only notice the compromise after seeing unusually rapid usage depletion. Anthropic has responded by revoking affected sessions, removing saved payment methods and promising refunds for unauthorized charges it identifies.

Looking Ahead

For Claude users, the immediate priority is to treat unexpected usage spikes as a possible security warning, particularly if a computer has recently installed software from an unofficial source. Users who suspect compromise should terminate unfamiliar Claude sessions, change relevant credentials and investigate their computers for infostealer malware. Anthropic’s active-session controls provide a way to review logged-in devices and remotely terminate suspicious sessions.

The broader lesson extends beyond Claude. As AI subscriptions and API access become valuable computing resources, attackers have a growing incentive to steal authenticated sessions and consume someone else’s AI capacity. Security providers may therefore need stronger anomaly detection, session protection and device-risk monitoring, while users will need to treat AI accounts with the same security discipline they apply to email, cloud storage and financial services.