Introducing GuardDog 3.0: A new rules engine, transparent sandboxing, and more
GuardDog 3.0, Datadog’s open-source malicious package detector, migrated its rule engine from Semgrep to YARA to improve performance and reduce memory overhead on large-scale scans. The release introduces a new risk-scoring engine that combines capability detection (what a package can do) and threat indicators (suspicious domains, deobfuscation routines, credential access patterns) to quantify maliciousness across a 0–10 scale based on attack chain completeness, specificity, and sophistication. GuardDog 3.0 also adds transparent sandboxing via nono-py, running extraction and analysis in an isolated environment with restricted network and filesystem access as defense-in-depth against vulnerabilities in the scanning tool itself. The tool now supports PyPI, npm, and Golang module scanning with a framework for detecting typical malware attack stages: initial access, post-compromise credential exfiltration, and persistence injection.