Hackers' data-stealing software is also grabbing AI coding tool logins

Malware that steals data from computers is now picking up the sign-in keys for AI coding tools, not just saved passwords. In an October 30 report, security firm SpyCloud said it found Claude Code credentials in logs recovered from an infostealer called WardenStealer. Its advice: when a computer is infected, check the AI tools too.

What SpyCloud found

An infostealer is malware that collects data from an infected machine and sends it away. SpyCloud studied logs from two families, WardenStealer and Aoki. In the WardenStealer logs, it found Claude Code's OAuth credentials file. That file holds access and refresh tokens. A token is like a key card: it lets software use an account without typing the password again.

The same file also held tokens for connected Model Context Protocol (MCP) servers. MCP is how AI tools link to other services. Those are separate keys, and each one needs its own check.

Keys versus context

SpyCloud also points to instruction files such as CLAUDE.md and MCP configuration. These are not keys. They describe the setup that a stolen token could reach into. Finding them in a log does not show that an attacker used them to break into another system.

How big is it

SpyCloud says it has ingested more than 130,000 unique WardenStealer infections since it first saw the malware on August 3, 2026. That number counts infections. It is not a count of hacked AI accounts. It also does not show how common AI-tool credential theft is among all developers.

What to do

SpyCloud says to "treat AI-tool credentials as in scope for the investigation." If a work computer is ever infected, its response list is:

• Rotate exposed keys and tokens, meaning replace them with new ones.
• Revoke active sessions so old logins stop working.
• Watch for unusual use of connected services and their APIs.

The point is to account for every kind of access the infected machine held. A new password alone is not a complete cleanup.