The AI Governance Gap Is Bigger Than We Think

This commentary reviews JFrog’s 2026 Software Supply Chain Security report and warns that while most organizations claim AI governance, governance often lags reality: repositories contain detected malicious payloads, malicious npm packages rose 451% with over 177,000 identified, JFrog found 495 malicious AI models and 969 malicious agent skills, and ~48,000 CVEs were disclosed in 2025. The author argues attackers are shifting upstream to exploit trust relationships inside development workflows (models, MCP servers, IDE extensions and agents), that vulnerability counts are a poor proxy for real risk, and that platform engineering must operationalize governance where AI and software delivery actually occur.