Executive brief
PraisonAI, an AI agent framework, contains a security flaw in its deployment module. An attacker with high-level access can inject malicious commands into the system's configuration files. If exploited, this allows the attacker to take full control of the server running the AI agents, potentially leading to data theft or complete service disruption.
Technical details
A code injection vulnerability exists in PraisonAI's deployment logic (specifically `deploy/api.py` and `deploy/docker.py`). The root cause is the direct interpolation of the `agents_file` parameter into Python f-strings used to generate server code and Dockerfiles. Because this input is not sanitized, an attacker with the ability to modify configuration or provide CLI arguments can break out of the string context to execute arbitrary Python code. The injected code is executed when the framework calls `subprocess.Popen()` to run the generated server file. This vulnerability requires high privileges (PR:H) but allows for a full scope escape (S:C), granting the attacker complete control over the underlying host. A fix is available in version 4.6.78.
Affected products
- MervinPraison PraisonAI < 4.6.78
Timeline
- 2026-06-25: advisory: Vendor published security advisory GHSA-g6j7-pffp-8whg
- 2026-07-10: disclosed: CVE-2026-61444 published to NVD