You open Claude, and your usage limit for the month is gone. Completely. You haven't logged in for days. No bug, no billing glitch — you probably had a digital visitor.
That's exactly what happened to several Claude users recently, and Anthropic has issued a warning about it: infostealer malware is stealing active login sessions and using them to chat on someone else's tab.
How the attack works
Infostealers aren't new. They run locally on infected machines and vacuum up everything the browser has stored — passwords, cookies, session tokens. That last one is the problem here. A session token essentially tells the server, "this person already logged in, no further check needed." Whoever steals that token needs neither your password nor your two-factor authentication (2FA) — they just walk in mid-session, as if they were you.
Anthropic names five malware families for Windows (Vidar, LummaC2, StealC, RedLine, Acreed) and one for Mac (Atomic Stealer, aka AMOS). According to Anthropic, only a small number of Mac users are affected; the campaign appears to be running wider on Windows.
How you'd notice
The most telling sign: your usage limit refills as usual at the start of the month — and then suddenly empties again, even though you haven't touched Claude. If that sounds familiar, get suspicious.
What Anthropic is doing
Anthropic is actively signing affected accounts out, removing saved payment methods, and refunding charges it identifies as unauthorized. The company stresses one thing explicitly: the malware has nothing to do with Claude itself and wasn't installed through Claude. It arrives the usual way — through shady downloads or fake apps — and simply harvests whatever it finds on the system. Claude accounts are just one of many targets.
That distinction matters for perspective: this isn't a data breach at Anthropic, and nobody hacked Claude itself. If your machine is infected, your email account, your online banking login, and a dozen other services with an active browser session are likely compromised too. Claude is one of many possible victims here — not the cause.
What you can actually do
Three steps, in the right order:
- Clean the machine first. Run a current virus scan (Windows Defender is a fine start; a second scanner like Malwarebytes doesn't hurt). Changing passwords before the malware is gone accomplishes nothing — the next login just gets intercepted again.
- Then change your passwords. Start with your email account, since it can reset almost everything else. Then Claude and any other service you're worried about.
- Check your billing. A look at your Claude usage history shows whether there was activity in the relevant period that wasn't you. If something looks off, contact Anthropic's support.
What about Linux?
Notable: there isn't a single Linux threat on Anthropic's list. Five times Windows, once Mac. That's no accident — but it's not to Linux's credit either.
Stealers are rented out as a service these days, and the loot is resold on criminal marketplaces within a day or two. At a desktop market share of around 5%, porting them simply doesn't pay off. So the advantage is real — but it's borrowed, and it rests entirely on how small the target group is.
Technically it's trivial, and it does happen. The remote-access trojan Chaos RAT recently made the rounds disguised as a "network analysis tool" for Linux. And in July 2025, three packages showed up in the Arch User Repository — Arch Linux's community package source — that installed exactly that trojan. They carried "fix" and "patched" in their names so they'd look like helpful corrections. They were online for two days before someone reported them.
And here's the genuinely uncomfortable part: the defenses against this specific attack are Windows-first. Chrome's extra layer of cookie encryption exists only there, and the new device-bound sessions landed there first. On Linux, the key sits in the desktop's keyring — and if none is running, in a passphrase hardcoded into the program. Anyone who can run code as your user reads your browser profile and your SSH keys without hitting another hurdle.
The attack path on Linux is just a different one. Not a fake EXE file, but a package from a community source, an npm install with a compromised dependency, a curl | bash from a guide nobody read. If you run Linux because you want control, this is the place to actually exercise it: take a moment to see what a script does before it runs as you.
The takeaway
Infostealers aren't a Claude-specific problem — they hit any service you leave logged in inside your browser. What you learn here applies just as much to ChatGPT, Gemini, or your email inbox. A current antivirus scanner and the habit of actually logging out on shared or public machines — not just closing the tab — protect you more than any single password change.
There's a small silver lining: Anthropic isn't staying quiet here, it's actively counter-acting — signing out, removing payment data, refunding. That doesn't replace your own caution, but it does mean an emptied usage quota isn't the end of the world — it's a fixable problem. Clean your machine once, and it's handled.