New AI Supply Chain Malware Steals Keys from Developer Pipelines While Hiding in Plain Sight

A self-propagating worm targeting AI development tools is spreading across software environments with limited detection, according to a CrowdStrike report on a strain called Sandworm_Mode. First identified by Socket in February 2026, the malware represents a growing class of supply-chain threats built specifically to exploit the complexity of modern AI-integrated development pipelines.

The worm targets sensitive data across the AI toolchain, including credentials, API keys for nine major LLM providers, and access tokens for CI/CD pipelines, cloud providers, and AI coding assistants. It spreads through code repositories and blends into the normal noise of AI-driven development environments, where pulling down and executing dependencies is a constant background activity.

Key Takeaways

• Sandworm_Mode steals credentials, keys, and secrets across the AI toolchain including 9 major LLM providers and CI/CD pipelines
• Sets multi-day delays between initial access and follow-on activity to widen the detection gap for defenders
• Destroys compromised environments if it cannot spread, indicating sophisticated and deliberate design
• CrowdStrike has not attributed the malware to a specific actor after four months of review; suspects nation-state or e-crime origins

"Trying to find the signal of something malicious happening is very difficult because there's so much noise out there," said Adam Meyers, senior vice president of counter adversary operations at CrowdStrike.

CrowdStrike continues to observe active supply-chain packages using similar but technically distinct techniques. Security teams embedded in AI development environments should treat dependency pulls as a new attack surface requiring active monitoring.

Read the full article on CyberScoop