PraisonAI is a multi-agent teams system. From 1.4.0 until 1.7.2, createAgentLoop() in src/praisonai-ts/src/ai/agent-loop.ts passes executable tools to generateText() before invoking the onToolCall approval callback. Because the wrapped AI SDK executes tool handlers during generation, a callback that returns false records tool_rejected only after the denied tool has already produced side effects and populated toolResults. Applications using onToolCall as a human or policy approval boundary can th

A security vulnerability (CVE-2026-57137) was discovered in PraisonAI, a multi-agent teams system, affecting versions 1.4.0 through 1.7.2. The flaw exists in the createAgentLoop() function within agent-loop.ts, where executable tools are passed to generateText() before the onToolCall approval callback is invoked. Because the underlying AI SDK executes tool handlers during generation, a rejection callback returning false only logs the rejection after the tool has already executed and produced side effects. This means applications relying on onToolCall as a human-in-the-loop or policy approval boundary are effectively bypassed. Attackers or unauthorized workflows could execute rejected file operations, system commands, API calls, or data-modifying operations without proper authorization. The vulnerability undermines trust boundaries in agentic AI systems where human oversight is critical. The issue has been patched in version 1.7.2 of PraisonAI.