Executive brief
CowAgent, an AI assistant and agent framework, contains a security flaw in its Vision Tool component. This tool is responsible for processing images, including fetching them from web addresses. An attacker can exploit this to force the server to make unauthorized requests to internal systems or private network services, potentially leading to the exposure of sensitive internal data or infrastructure details.
Technical details
A Server-Side Request Forgery (SSRF) vulnerability exists in CowAgent up to version 2.1.1 within the Vision Tool component. The functions `_build_image_content` and `_download_to_data_url` in `agent/tools/vision/vision.py` fail to validate user-supplied image URLs before fetching them. A remote, unauthenticated attacker can provide a URL pointing to internal loopback addresses (e.g., 127.0.0.1) or private network ranges (RFC1918) to probe internal services or cloud metadata endpoints. The issue is resolved in version 2.1.2 by implementing DNS resolution and IP range validation in a new `_validate_url_safe()` function.
Affected products
- zhayujie CowAgent up to 2.1.1
Timeline
- 2026-06-09: disclosed: Issue reported on GitHub
- 2026-06-11: patched: Fix committed to repository
- 2026-07-10: advisory: CVE published to NVD
References
- https://github.com/zhayujie/CowAgent/
- https://github.com/zhayujie/CowAgent/commit/e85290cddcbb5ffc9c235927f4c92e5b4c3ec264
- https://github.com/zhayujie/CowAgent/issues/2872
- https://github.com/zhayujie/CowAgent/pull/2886
- https://github.com/zhayujie/CowAgent/releases/tag/2.1.2
- https://vuldb.com/cve/CVE-2026-15330
- https://vuldb.com/submit/853103