Executive brief
CowAgent is an AI assistant that can perform tasks and use tools like web browsing. A security flaw in its web-fetching tool allows the AI (or someone influencing it via prompt injection) to access internal network resources, such as private company servers or cloud metadata services. This could lead to the exposure of sensitive internal data or unauthorized access to local infrastructure. Users should upgrade to version 2.1.2 to resolve this issue.
Technical details
A Server-Side Request Forgery (SSRF) vulnerability exists in CowAgent's 'web_fetch' tool within the WebFetch.execute function in agent/tools/web_fetch/web_fetch.py. The component failed to validate model-supplied URLs, allowing requests to loopback addresses (127.0.0.1), RFC1918 private networks, and cloud metadata endpoints (169.254.169.254). Additionally, the tool followed HTTP redirects without re-validating the target, enabling attackers to bypass initial checks via a 302 redirect from a public to a private address. An attacker can exploit this via remote network access, potentially through prompt injection, to exfiltrate internal service data. The issue is fixed in version 2.1.2 by implementing a shared URL safety validator and manual redirect following with hop-by-hop validation.
Affected products
- zhayujie CowAgent up to 2.1.1
Timeline
- 2026-06-17: patched: Patch ea47f3097eed4f8295c4cb3d76ecb97e0f43d632 committed
- 2026-07-18: disclosed: CVE-2026-16194 published
References
- https://github.com/zhayujie/CowAgent/
- https://github.com/zhayujie/CowAgent/commit/ea47f3097eed4f8295c4cb3d76ecb97e0f43d632
- https://github.com/zhayujie/CowAgent/issues/2889
- https://github.com/zhayujie/CowAgent/pull/2900
- https://github.com/zhayujie/CowAgent/releases/tag/2.1.2
- https://vuldb.com/cve/CVE-2026-16194
- https://vuldb.com/submit/857620