As artificial intelligence tools become standard in software development, cybersecurity firm CrowdStrike has uncovered a worm that targets the AI toolchain to steal credentials, exfiltrate data, and destroy systems—all while masquerading as normal operations. The malware, discovered during investigations into AI software supply chain attacks, represents what researchers call an emerging class of threats. Adam Meyers, CrowdStrike's senior vice president of counter adversary work, says the company hasn't attributed the activity to a specific group but notes it aligns with tactics used by actors like TeamPCP (tracked as “Altered Spider”) and North Korean hackers.

The worm operates in phases, starting with reconnaissance of the target environment. It then hunts for access tokens, cryptographic keys, and server credentials, especially npm tokens that unlock software package management servers and development capabilities like pull requests. As it gains privileges, the malware unpacks further and can activate a “death switch” to destroy files or block legitimate access. The critical insight: much of this malicious activity happens in blind spots because the worm mimics legitimate automation tools used for building code. “It's like a needle in a haystack, except this is a needle in a needle stack,” Meyers says.

Detection is notoriously difficult because AI coding systems generate telemetry that overlaps heavily with the worm's behavior. Attackers also embed time delays, executing capabilities hours or days after initial access, making it hard for defenders to trace cause and effect. Meyers emphasizes that as AI software development explodes, the industry must collaborate on structural solutions. “It’s a limited detection surface because only so much of this activity produces any sort of telemetry signal for us to look at,” he says, “so it becomes extremely onerous to determine what is legitimate and what is illegitimate behavior.”

Source: Wired