Executive brief
zhayujie chatgpt-on-wechat CowAgent is an AI agent framework that integrates ChatGPT with the WeChat messaging platform. A security flaw in the WeChat Official Account endpoint allows remote attackers to bypass authentication if the system is not configured with a specific security token. This could allow unauthorized parties to interact with the webhook endpoint, potentially disrupting service or sending unauthorized commands.
Technical details
An authentication bypass vulnerability exists in the `verify_server` function within `channel/wechatmp/common.py` of CowAgent 2.1.0. The root cause is a failure to check if the `wechatmp_token` is empty or unconfigured; when missing, the signature verification process degenerates into a predictable hash based on attacker-controlled parameters (timestamp, nonce, etc.). A remote, unauthenticated attacker can exploit this to successfully pass the server verification check on the `/wx` endpoint. The issue is resolved in version 2.1.1 by implementing an explicit non-empty check that returns a 403 Forbidden error if the token is missing.
Affected products
- zhayujie chatgpt-on-wechat CowAgent 2.1.0
Timeline
- 2026-06-09: patched: Version 2.1.1 released with fix
- 2026-07-05: disclosed: CVE-2026-14714 published
References
- https://github.com/zhayujie/CowAgent/commit/3d7c68bac6ee74fad63f43cf99e45c62e202ed55
- https://github.com/zhayujie/CowAgent/issues/2860
- https://github.com/zhayujie/CowAgent/releases/tag/2.1.1
- https://vuldb.com/cve/CVE-2026-14714
- https://vuldb.com/submit/847484
- https://vuldb.com/vuln/376304
- https://vuldb.com/vuln/376304/cti