OpenClaw affected by SSRF in Image Tool Remote Fetch

Summary

A server-side request forgery (SSRF) vulnerability in the Image tool allowed attackers to force OpenClaw to make HTTP requests to arbitrary internal or restricted network targets.

Affected Versions

• npm: openclaw <= 2026.2.1

Patched Versions

• npm: openclaw 2026.2.2 and later

Fix Commits

• 81c68f582d4a9a20d9cca9f367d2da9edc5a65ae (guard remote media fetches with SSRF checks)
• 9bd64c8a1f91dda602afc1d5246a2ff2be164647 (expand SSRF guard coverage)

Details

The Image tool accepts file paths, file:// URLs, data: URLs, and http(s) URLs. In vulnerable versions, http(s) URLs were fetched without SSRF protections, enabling requests to localhost, RFC1918, link-local, and cloud metadata targets.

This was fixed by routing remote media fetching through the SSRF guard (private/internal IP + hostname blocking, redirect hardening, DNS pinning).

Exploitability Notes

• Requires attacker-controlled invocation of the Image tool (direct tool access, or a gateway/channel surface that forwards untrusted image arguments into tool calls).
• The image tool expects the fetched content to be an image. Many high-value SSRF targets return text/JSON (for example cloud metadata endpoints), which will typically fail media-type validation. In practice, the most direct confidentiality impact comes from internal endpoints that actually return images (screenshots/renderers, camera snapshots, chart exports, etc.).
• Remote fetches are GET-only with no custom headers. Some metadata services require special headers or session tokens (for example GCP Metadata-Flavor, AWS IMDSv2 token), which can further reduce the likelihood of direct credential theft in some environments.
• Despite the above constraints, SSRF remains a powerful primitive: it can enable internal network probing and access to unauthenticated/internal HTTP endpoints, and can chain with other weaknesses if present.

Thanks @p80n-sec for reporting.