An unsafe reflection vulnerability in MervinPraison PraisonAI and praisonaiagents allows unauthenticated remote attackers to invoke arbitrary application callables, leading to potential integrity compromise and partial data access.

This is an unsafe reflection flaw classified as CWE-470, where externally controlled input selects classes or code. The application resolves unresolved tool names against module globals and main when default permissions do not reject undeclared tool names, requiring no authentication.

The high CVSS score of 8.6 reflects significant potential for unauthorized control execution and data tampering. A successful exploit could allow attackers to execute unintended functions within the application context, resulting in system compromise, unauthorized data modification, and potential operational disruption.

Immediate Action: Update praisonai to version 4.6.37 and praisonaiagents to version 1.6.37 or later via PyPI.

Proactive Monitoring: Monitor application logs for unusual tool invocation patterns, unexpected module calls, or unauthorized interaction attempts with internal agent functions.

Compensating Controls: Implement strict input validation rules at the network perimeter or application gateway to inspect tool-call parameters and reject unexpected or malformed inputs.

Public Exploit Available: No

Given the high severity score and the availability of a proof-of-concept, organizations must prioritize patching immediately. Administrators should upgrade both praisonai and praisonaiagents to their respective fixed versions to eliminate the reflection vector and secure the agent environment.