Vulnerability  ·  2026-06-20

PraisonAI AGUI Endpoint Unauthenticated Cross-Origin Agent Execution (CVSS 8.1)

VulnerabilityHigh impactGlobalCVE-2026-56076
PraisonAI's POST /agui endpoint for the AG-UI protocol lacks any authentication and hardcodes a wildcard CORS policy (Access-Control-Allow-Origin: *). Because Starlette parses JSON regardless of Content-Type, attackers can craft simple cross-origin requests that bypass CORS preflight entirely, remotely triggering arbitrary agent execution and receiving sensitive response data. Published 2026-06-18, CVSS 8.1 HIGH.
Unauthenticated remote agent invocation is a critical agentic attack: any website can silently trigger agent runs against a victim's PraisonAI instance via cross-site request, exfiltrating tool execution results and environment variables without the user's knowledge. This is a real-world implementation of the cross-origin agent hijack attack class.
Remote attacker sends a crafted cross-origin POST request to the /agui endpoint. The endpoint lacks authentication and hardcodes Access-Control-Allow-Origin: * headers. Combined with Starlette's Content-Type-agnostic JSON parsing, attackers can bypass CORS preflight via simple requests (no preflight triggered), causing arbitrary agent execution and exfiltrating tool output including environment secrets.
PraisonAI < 1.5.128
Upgrade to PraisonAI 1.5.128 or later. Advisory: https://github.com/MervinPraison/PraisonAI/security/advisories/GHSA-x462-jjpc-q4q4
Sources
NVD CVE-2026-56076VulnCheck Advisory — PraisonAI cross-origin agent executionGitHub Advisory GHSA-x462-jjpc-q4q4
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →