Hackers Are Quietly Draining Claude Users’ AI Tokens

Hackers Are Quietly Draining Claude Users’ AI Tokens

Imagine paying $200 every month for Claude, barely using it, and then watching your AI allowance disappear.

That is exactly what happened to one Claude subscriber in the UK.

And the worrying part? He was not even using Claude when the tokens were being consumed.

The strange Claude activity

Grant De Swardt, an independent AI consultant in East Sussex, first noticed something unusual on August 4. His Claude Max 20x account was using tokens even though he was not working.

The next day, he decided to test it properly. He disabled the tools connected to Claude, stopped working, and made sure his scheduled tasks were paused.

Still, the usage climbed.

During one period, his allowance jumped from 45% to 55% while he was doing nothing. At that point, this was no longer just a strange usage graph.

Something or someone appeared to be using his account.

Then Anthropic found the problem

De Swardt contacted Anthropic and asked for a detailed breakdown of what was consuming his tokens. The company could not provide an itemized report, but it agreed that something was wrong.

Anthropic suspended his account, invalidated his sessions and Claude Code tokens, and refunded him £44.49 for the unused portion of his $200 subscription. After investigating, Anthropic reportedly told him that a compromised Claude session key had been used to create unauthorized Claude Code access tokens.

In simple terms, someone had gained access to his logged-in Claude session and was using his account's AI capacity.

And they were not necessarily using it for him.

He was not the only one

Things became even more concerning when De Swardt shared his experience on Reddit.

More than 80 people responded with similar stories.

Some reported their usage suddenly jumping from almost nothing to huge amounts. One user claimed their account reached 100% usage without them using it. Another reported going from 0% to 49% in just 12 minutes.

Another user said their maximum allowance was consumed for three consecutive days while they were not using Claude. That suggests this may be bigger than one compromised account.

Anthropic points to malware

Anthropic later warned some affected users that it had identified a bad actor using infostealer malware.

Infostealers are malicious programs designed to steal things such as passwords and active login sessions from someone's computer. The stolen sessions can then potentially be used to access online accounts without the attacker needing the user's password again.

Anthropic said the malware did not come from using Claude itself. It can enter computers through infected software, malicious advertisements and other online sources. The company said that when it detects suspicious activity, it can sign users out, invalidate authorizations, and provide refunds in some cases.

But there is still a major problem.

Where did the tokens go?

Users currently do not have a detailed usage history showing exactly what consumed their Claude allowance. That makes this type of theft difficult to spot.

A user might simply see their monthly allowance falling and assume Claude is being used heavily in the background. By the time they realize something is wrong, a significant amount of their paid AI capacity could already be gone.

De Swardt's case also highlights how serious that can become for people who depend on AI for their businesses.

His consultancy uses AI agents for administration, coding, website work and automating business processes.

So when Anthropic suspended his account during its investigation, it did not just interrupt an AI subscription.

It disrupted his business.

After roughly two weeks, his account was restored. But he eventually cancelled Claude and moved to Cursor, saying he remained concerned about the lack of visibility into what was consuming his tokens.

The bigger lesson

AI subscriptions are becoming more than simple chat apps.

People are connecting them to coding environments, automation tools, business systems, and AI agents. That means a stolen session can potentially be much more valuable than a normal account password.

And this incident raises an uncomfortable question:

If someone secretly uses your AI account, how quickly would you actually know?

For Claude users, the safest move is to take unusual usage seriously, keep computers protected, avoid suspicious downloads and ads, and review connected services and active sessions regularly.

Because in the new AI economy, hackers may not always want your money.

They may want your AI.