Cybersecurity researchers have uncovered a sophisticated new worm designed to infiltrate AI development pipelines, steal sensitive credentials, and even sabotage systems. The malware cleverly disguises its malicious activities as legitimate operations, making it incredibly hard to detect.
CrowdStrike researchers identified the worm while investigating attacks targeting the AI software supply chain. While the specific culprit remains unconfirmed, the tactics align with evolving threats from actors like TeamPCP (tracked as "Altered Spider") and North Korean groups. This discovery highlights a new class of attack specifically targeting the growing reliance on AI in software development.
"This is one of the campaigns that we’ve seen showing that this is an emerging attack class," said Adam Meyers, CrowdStrike's senior vice president of counter adversary work. "As AI coding agents become the development standard, supply chain threats are evolving to exploit those trust relationships."
The worm operates in stages, starting with reconnaissance to map out the target environment. It then hunts for access tokens, cryptographic keys, and server credentials. As it gains more privileges, it can access critical tools like "npm" for software package management and even manipulate pull requests. Deeper infiltration allows it to exfiltrate more data and potentially deploy a "death switch" to destroy files or lock down compromised infrastructure.
The biggest challenge in detecting this threat, according to Meyers, is its ability to mimic legitimate automation processes used in code building. "It's like a needle in a haystack, except this is a needle in a needle stack," he explained. "This looks very much like a lot of the automation organizations are using to build code, so it’s very difficult to detect."
Adding to the stealth, the worm incorporates deliberate time delays, executing malicious actions hours or even days after initial compromise, obscuring the chain of events. Meyers stressed the urgent need for collaboration among industry players to develop structural solutions as AI development continues to explode, noting that the limited telemetry signals make distinguishing legitimate from illegitimate behavior a significant hurdle.
