Claude’s Token Theft and Hidden Tier Caps Create Risk

Power users and businesses relying on Anthropic’s Claude models are encountering significant friction on two fronts. Recent reports from The Verge and TechCrunch highlight legal disputes regarding subscription limits alongside security vulnerabilities that allow unauthorized token consumption.

For teams integrating AI into daily operations, these issues combine financial unpredictability with operational exposure. Understanding how subscription terms work and how session tokens can be compromised is now critical for maintaining secure, cost-effective workflows.

Why are Claude subscribers taking Anthropic to court?

An expanded class action lawsuit filed by former FTC attorneys Monica Vaca and Kati Daffan accuses Anthropic of deceptively advertising its high-tier subscriptions. The lawsuit targets the Max plan, which costs between $100 and $200 per month and promises 5x to 20x the usage allowance of the standard $20 Pro plan.

According to the complaint covered by The Verge, users who upgrade expect a proportional increase in total weekly capacity. Instead, the marketing graphics omit key constraints that require navigating multiple hyperlinks to discover. The 5x or 20x capacity boost only applies to rolling 5-hour windows, which are themselves bounded by a strict weekly allowance cap.

"The rolling 5-hour window is what actually controls whether you can work... It’s like giving someone a bigger gas tank while keeping the fuel pump limited to one gallon every five hours."

When users hit these hidden limits during heavy workloads, their workflow stops cold despite paying premium rates.

How are hackers consuming subscriber tokens without permission?

While lawsuit allegations center on advertising opacity, a parallel issue reported by TechCrunch demonstrates technical vulnerabilities in session management. Attackers using standard infostealer malware are harvesting Claude session keys from user devices, using them to mint unauthorized Claude Code OAuth tokens.

These stolen tokens allow unauthorized third parties to drain a subscriber's usage quota. In one documented instance, independent consultant Grant De Swardt observed his token usage increase from 45% to 55% while his workstation was powered down and all automated scripts were paused.

Anthropic confirmed that infostealer malware was leveraging valid session data to run tasks for outside parties. However, because Anthropic does not provide itemized usage logs to account holders, detecting token theft requires manual monitoring of quota percentages.

What is the combined operational impact on businesses?

The intersection of opaque limits and session hijacking creates severe operational risks for professionals and small teams:

• Sudden Account Suspensions: When Anthropic detects suspicious token usage, it routinely invalidates sessions and suspends accounts. For consultants relying on AI agents for daily operations, a multi-week suspension disrupts client deliverables.
• Inability to Audit Spend: Without granular usage logs detailing API calls, timestamps, or IP addresses, account managers cannot verify whether a quota depletion stems from normal team activity, background agents, hidden session limits, or active theft.
• Inadequate Financial Redress: Affected users report receiving partial refunds—such as £44.49 on a $200 monthly tier—which fail to offset the business disruption caused by locked access.

How can management secure Claude credentials on Monday morning?

To mitigate security and financial risks, operational leaders should implement immediate security hygiene and contractual reviews.

First, mandate that all team members explicitly log out of web sessions and rotate API keys regularly. Session tokens stored in local browser storage remain vulnerable if local machines contract infostealer malware from unrelated software downloads or malicious web ads.

Second, establish dedicated monitoring for quota tracking. Assign a team member to verify usage metrics daily; if consumption spikes during zero-activity windows, immediately revoke all active OAuth authorizations and contact enterprise support.

Finally, evaluate whether high-tier subscriptions match actual concurrency needs. If session-based throttling restricts your throughput despite paying for maximum capacity, multi-model execution tools or direct API integrations with strict budget caps may provide greater reliability and control.