Hackers Exploit Stolen Session Keys to Drain Claude AI Token Credits

A months-long account theft campaign has targeted Claude AI subscribers, with attackers draining paid token allowances through compromised login sessions. The scheme went undetected for extended periods because Anthropic's billing dashboard lacks itemized usage tracking.

How the Attack Works#

According to an official Anthropic communication, "a bad actor 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." Infostealers are malicious programs that harvest saved passwords, session cookies, and authentication credentials stored on infected devices.

Once hackers obtain a valid session key, they can:

• Create unauthorized Claude Code OAuth tokens
• Access and use the account without the owner's knowledge
• Drain monthly token allowances silently

The Pattern Emerges#

In early August, Grant de Swardt, an AI consultant based in East Sussex, UK, discovered his Claude Max 20x account was rapidly consuming tokens despite his not using the service. Over a controlled test period, his usage jumped from 45% to 55% while all scheduled tasks remained paused and no active work occurred.

After contacting Anthropic, the company suspended his account and refunded £44.49 for unused time. Following its investigation, Anthropic concluded that a compromised session key had been exploited to mint unauthorized tokens—though it could not pinpoint whether credentials were stolen from his machine or the account had been linked to a hostile third-party service.

De Swardt's post on Reddit sparked recognition from dozens of other users reporting identical symptoms:

• One subscriber watched their account auto-upgrade without consent and tokens jump from 0% to 100% instantly
• Another saw 49% consumption in 12 minutes despite only running a couple of prompts
• A third user's account burned through its full monthly allowance for three consecutive days with zero activity

Some affected users shared warning emails from Anthropic confirming token theft and recommending computer scans for malware.

A Structural Vulnerability#

The theft highlights a critical gap in Claude's user protections: Anthropic's account dashboard tracks only aggregate token usage, not itemized activity logs. Users cannot see which API calls, sessions, or operations consumed tokens—even when requesting an itemized breakdown.