critical • SecOpsAI intelligence

Mini Shai-Hulud crosses npm and PyPI: advisory protection for removed artifacts

Critical Published: 2026-05-12T00:00:00Z Updated: 2026-05-12T00:00:00Z
Supply Chain Advisories Detection Engineering Mitigation

Mini Shai-Hulud crosses npm and PyPI: advisory protection for removed artifacts

Executive Summary

Mini Shai-Hulud is a confirmed software supply-chain campaign affecting npm and PyPI packages. Some compromised versions were removed from public registries quickly, which is good for users but creates a scanner blind spot: artifact diffing can fail after takedown.

SecOpsAI now ships an emergency advisory ingestion path. Named compromised versions can produce high-confidence SOC findings even when the malicious artifact is no longer fetchable.

Affected Artifacts

What SecOpsAI Detected

Local SecOpsAI findings already identified suspicious behavior in mistralai@2.4.6, including subprocess execution, shell downloader behavior, network egress, artifact divergence, and suspicious code present in only one PyPI artifact path.

What Was Missed Before Advisory Ingestion

Removed npm/PyPI artifacts such as some @opensearch-project/opensearch, guardrails-ai, and @squawk/* versions could previously end as diff generation failed. That error was technically true, but operationally weak: the version was still confirmed compromised by external reporting.

New Protection

Emergency advisories are stored as source-backed JSON under data/advisories/. The scanner checks advisories before allowlist or reputation shortcuts. If a diff succeeds, the advisory enriches the finding. If artifact fetch or diff generation fails, the advisory still creates a malicious high-confidence SOC finding.

IOCs And Behaviors

Detection Logic

secopsai supply-chain advisory check --ecosystem npm --package @opensearch-project/opensearch --version 3.8.0
secopsai supply-chain explain-verdict --ecosystem pypi --package guardrails-ai --version 0.10.1
secopsai supply-chain reconcile-history --include-advisories

Recommended Actions

Timeline

References

Comments

Comments are moderated before publication. Do not post secrets, tokens, customer data, or exploit payloads.