Executive brief
PraisonAI is a framework used to build and manage autonomous AI agents. A security flaw in its monitoring component allows an attacker to manipulate file paths by providing specially crafted agent IDs. This could allow an unauthorized user to read sensitive system files, delete data, or potentially take control of the server.
Technical details
A path traversal vulnerability (CWE-22) exists in the MultiAgentMonitor component of PraisonAI. The root cause is the lack of sanitization of agent IDs when they are concatenated with a base directory to create file paths. An authenticated attacker with network access can provide an agent ID containing traversal sequences (e.g., '../') to read, write, or overwrite files outside the intended directory. This can lead to sensitive information disclosure, denial of service via file corruption, or remote code execution if executable files are overwritten. The issue is fixed in version 1.5.115. Additionally, the advisory notes a related memory state leakage issue where non-unique agent IDs allow context sharing between different agents.
Affected products
- MervinPraison PraisonAI (praisonaiagents) < 1.5.115
Timeline
- 2026-04-07: advisory: Initial GitHub security advisory published
- 2026-06-18: disclosed: CVE-2026-56078 published to NVD
- 2026-06-18: patched: Fix confirmed in version 1.5.115