Executive brief
PraisonAI, a platform for managing AI agents, contains a security flaw that allows unauthorized users to bypass authentication when a specific 'disabled' mode is active. Although this mode was intended to only work for local users on the same machine, an attacker can trick the system into thinking they are local by sending a fake network header. This allows an attacker to list and trigger AI agents without permission, potentially leading to unauthorized data processing or service misuse.
Technical details
An authentication bypass exists in the Call API agent invocation endpoints (src/praisonai/praisonai/api/agent_invoke.py) of PraisonAI when the PRAISONAI_CALL_AUTH=disabled configuration is used. The application attempts to restrict this unauthenticated mode to localhost by checking request.url.hostname; however, this value is derived from the client-controlled HTTP Host header. A remote, unauthenticated attacker can send a request with a 'Host: 127.0.0.1' header to bypass this check. Successful exploitation allows the attacker to list registered agents via GET /api/v1/agents and invoke them via POST /api/v1/agents/{agent_id}/invoke. The issue is addressed in version 4.6.78.
Affected products
- MervinPraison PraisonAI < 4.6.78
Timeline
- 2026-06-25: advisory: GitHub Security Advisory GHSA-2gpf-2492-q9jh published
- 2026-07-15: disclosed: CVE-2026-61435 published to NVD
References
- https://github.com/MervinPraison/PraisonAI/commit/2a855c470077c7d2e2479a575f7ef7f548d51c33
- https://github.com/MervinPraison/PraisonAI/commit/846568c7a5d8ce9e71e56e4c213f027c04909753
- https://github.com/MervinPraison/PraisonAI/security/advisories/GHSA-2gpf-2492-q9jh
- https://www.vulncheck.com/advisories/praisonai-before-authentication-bypass-via-host-header-spoofing