GPU VulnDB

Database/AI/ML frameworks & serving

Crawl4AI: file:// URLs accepted by four endpoints let an unauthenticated caller read any file on the host

CVSS 9.2CVE-2026-26217AI/ML frameworks & servingcurated

Impact

The /execute_js, /screenshot, /pdf and /html endpoints accept file:// URLs, so an unauthenticated caller turns the crawler into an arbitrary file reader against its own container. The advisory names /etc/passwd, /etc/shadow, application config and /proc/self/environ - which is where the API keys and model-provider credentials of an ingestion service live. This is a credential-theft primitive for the pipeline Crawl4AI feeds rather than code execution; it is a separate flaw from the hooks exec() issue (CVE-2026-26216) with its own advisory, though the same 0.8.0 release fixes both.

Who can reach it

Any unauthenticated client that can reach the Crawl4AI Docker API HTTP endpoint. No credentials required.

What to do

Upgrade to Crawl4AI 0.8.0 or later and redeploy the container - image pull plus daemon restart, no node drain. Assume any secret that was present in the container environment is disclosed if the endpoint was reachable from untrusted networks, and rotate it.

References

Related entries

All AI/ML frameworks & serving entries

This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.