CrowdStrike Finds Self-Hiding Worm Targeting AI Coding Toolchains

Cybersecurity firm CrowdStrike says it discovered a worm actively targeting AI software development infrastructure while investigating supply chain attacks. Adam Meyers, the company's senior vice president of counter adversary operations, told WIRED the activity has not yet been attributed to a specific actor, but fits a broader pattern of groups such as TeamPCP (tracked by CrowdStrike as "Altered Spider") and North Korean hacking groups increasingly targeting the AI toolchain.

The worm operates in stages: it first reconnoiters the target environment, then hunts for access tokens, cryptographic keys and server credentials. As it escalates privileges, it continues harvesting data, with particular focus on npm tokens that unlock package-management servers and other development capabilities such as pull requests. Once deep enough inside a system, the malware can activate what Meyers calls a "death switch," destroying files or locking out legitimate users.

The worm's most dangerous trait is how well it blends in: much of its behavior closely mirrors the automation traffic that legitimate AI coding tools generate, making it extremely hard for security tools to distinguish malicious activity from normal telemetry. Its authors also built in delays of hours or even days between stages, further complicating efforts to trace cause and effect. CrowdStrike says it is developing better detection methods but stresses that defending AI development pipelines will require broader industry collaboration, given how limited the available detection signals currently are.