Credential Theft Surpasses Cryptomining as Dominant Open-Source Malware Threat

Open-source malware is undergoing a fundamental transformation, moving from resource abuse to access theft. Research indicates that over half of malicious packages now focus on stealing credentials and secrets, displacing cryptomining as the primary threat. This shift reflects an attacker strategy that prioritizes long-term, hard-to-detect access over noisy, short-lived CPU hijacking.

The numbers signal industrialized campaigns. Dropper and loader behavior surged by nearly 2900% year over year in Q1 2025, and around 77% of malicious packages now bundle multiple threat types. Droppers appear in almost every observed case, while secret exfiltration features in close to two-thirds. These packages are designed to evolve after installation, fetching additional payloads or altering behavior to avoid detection.

Credentials offer greater value than compute cycles. A stolen API key or token can unlock entire systems, spread across environments, and persist without raising resource alerts. By contrast, cryptomining triggers performance alarms and is quickly shut down. Attackers are exploiting the trust placed in developer workflows, embedding themselves inside build pipelines where automation can hide malicious activity.

Nowhere is the concentration risk more acute than in the JavaScript ecosystem. Modern applications often rely on hundreds of direct and transitive npm packages. Activity attributed to the Lazarus group illustrates the danger: over 200 malicious packages were identified, almost all concentrated in npm. Because the same ecosystem underpins financial services, government platforms, and critical national infrastructure, a single compromised dependency can propagate through shared frameworks and CI pipelines, turning an upstream flaw into a widespread downstream compromise.

CI/CD pipelines, built for speed and automation, become amplifiers. A tainted dependency can extract API keys, certificates, and tokens as part of routine builds, spreading across hundreds or thousands of deployments before anyone notices. The security boundary can no longer stop at runtime; it must encompass dependency intake and the developer environment.

AI-assisted coding introduces a fresh risk. Studies show that large language models sometimes hallucinate package names, tempting developers to install nonexistent dependencies. Attackers are already seeding packages with names that mimic these hallucinations, widening the attack surface. The author’s research found that smaller AI models, when augmented with real-time, authoritative package intelligence, significantly outperform larger standalone models on dependency tasks. This suggests that grounding AI tools in live ecosystem data is more critical than model size for secure software supply chains.

To realign defenses with upstream threats, UK organizations should gate dependency intake with automated policy enforcement, continuously monitor build environments for secret exposure, enforce provenance and integrity checks across all components, and ground AI coding assistants in verified package intelligence. These measures represent the practical meaning of “shift left”: blocking malicious code before it enters the build graph, not just reacting after deployment.

The quiet shift from coins to credentials has already taken place. The question is whether defenses will adapt quickly enough to match an adversary that now operates upstream, automated, and inside the supply chain.