Hackers Stole Claude AI Tokens From Paying Subscribers, and Anthropic Couldn't Tell Users What Was Happening
A British AI consultant watched his $200-a-month Claude account drain on days he never touched it. He was not the only one.
Key points
- On August 4, Grant de Swardt, an independent AI consultant in East Sussex, UK, noticed his Claude Max account consuming tokens on a day he had not worked.
- A compromised session key, the stored login credential that keeps you signed in, was used to create unauthorised access tokens and funnel his usage to unknown third parties.
- Some affected users received warning emails from Anthropic; de Swardt did not, and only found the wider pattern after posting on Reddit.
- Anthropic refunded de Swardt £44.49 and suspended his account; he cancelled his $200-a-month subscription after roughly two weeks without a full explanation.
- Anthropic provides no itemised usage log, meaning token theft of this kind could run undetected for months.
Grant de Swardt runs a one-person business in East Sussex helping small companies set up AI agents, software that carries out multi-step tasks on its own, for things like pulling purchase-order data from emails into accounting software. Claude, Anthropic's AI assistant, powers most of what he does daily: admin, website work, coding. When his Claude Max 20x subscription started burning through its token allowance (a token is roughly a word or part of a word; your plan gives you a fixed number before you hit a limit) on days he was not working, he noticed.
On August 5 he disabled every connected tool and ran no tasks himself. His usage still climbed from 45 percent to 55 percent with nothing running.
He contacted Anthropic and asked for an itemised breakdown of what was consuming his tokens. Anthropic could not provide one. It did agree something was wrong: it suspended his account, invalidated all active sessions, and issued a partial refund of £44.49 for the remaining days on his plan.
What actually happened to his account?
Anthropic told de Swardt that a compromised Claude session key, the stored credential that keeps you logged in without re-entering your password, had been used to create unauthorised OAuth tokens. OAuth tokens are short-lived digital passes that let one application access your account on its behalf. Someone had used his login to generate those passes and was quietly running their own activity through his account. Anthropic said it couldn't determine how access was first obtained, telling him the evidence was consistent with credentials being taken without his knowledge or with the account having been connected to an outside service.
Anthropics email to other affected users, first reported by TechCrunch, named the method: infostealer malware. Infostealers are malicious programs that install silently on a computer, then copy saved passwords and session data. That malware did not come from Claude itself; it can arrive through downloaded software, infected adverts or cracked apps.
The email read: "We have recently become aware of a bad actor that 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."
De Swardt says he found no evidence of malware on his machine and still doesn't know how access was obtained.
Was this just one person?
No. After posting on Reddit, de Swardt collected more than 80 replies from users describing similar patterns: accounts upgraded without consent, usage jumping from 0 to 49 percent in 12 minutes after a couple of prompts, or hitting the daily maximum for three consecutive days with no activity at all. Our September 8 story on Anthropic's disputed Claude subscription limits shows that questions about what subscribers actually receive for their money have been building for weeks.
Anthropics response varied. Some users received proactive warning emails. De Swardt did not.
What should Claude subscribers do right now?
Anthropics current dashboard does not show a line-by-line breakdown of what consumed your tokens. Without it, you can't tell whether unusual usage is your own or someone else's. That's the core problem, and Anthropic has not said when or whether it plans to fix it.
Four practical steps can reduce your exposure. Check active sessions in your Anthropic account settings and revoke any you don't recognise. Sign out and sign back in to invalidate old session tokens. Run a malware scan on any computer you use to access Claude. Watch your usage bar regularly; if it moves on days you haven't worked, contact Anthropic support and request an account review.
De Swardt's account was reinstated after about two weeks. He cancelled anyway, switching to Cursor, a coding tool that connects to several AI models including cheaper open-source options. "I can't see going back," he told TechCrunch, "without them actually having resolved the issue in any way." Anthropic declined to comment on what tools users can use to spot misuse.
The plain judgement here: this isn't a story about a single stolen account. It's about a billing system with no receipt. Until Anthropic gives subscribers a line-by-line log, anyone on a paid plan is trusting a total on a meter they can't read.
Common questions
Can I see exactly what is using my Claude tokens?
Not currently. Anthropic's dashboard shows total usage but no itemised log, so distinguishing your own activity from unauthorised access requires contacting support directly.
Does this mean Anthropic's systems were hacked?
Anthropics own infrastructure does not appear to have been breached. The attack worked by stealing login sessions from users' computers using widely available malware, then using those sessions to access accounts normally.
