Vulnerability in PraisonAI Allows Unauthorized HTTP Requests by Exploiting Allowlist Bypass

CVE-2026-55532

What is CVE-2026-55532?

PraisonAI, a multi-agent teams system, contains a vulnerability where the MCP HTTP Stream _validate_origin method improperly allows requests from attacker-controlled origins, such as localhost.attacker.com, to bypass the allowed origins list. This permits webpages to send Content-Type: text/plain requests without preflight checks. Consequently, attackers can invoke certain tools and execute file writes that persist agent instructions, posing a significant risk to system integrity. The issue has been resolved in version 4.6.58.

References