FastGPT Critical Vulnerability CVE-2026-42302 Enables Unauthenticated RCE via Port 8080

A critical security vulnerability has been disclosed in FastGPT, affecting versions 4.14.10 through 4.14.12. Tracked as CVE-2026-42302, the flaw allows unauthenticated remote code execution through the agent-sandbox component exposed on port 8080. The vulnerability is classified as CWE-306, indicating missing authentication for a critical function—meaning attackers can exploit the service without any credentials.

The affected component, agent-sandbox, is designed to execute code in an isolated environment, making this vulnerability particularly dangerous. By sending crafted requests to port 8080, an unauthenticated attacker could achieve arbitrary code execution on the underlying system. Organizations running FastGPT versions 4.14.10, 4.14.11, or 4.14.12 with the agent-sandbox service exposed to networks should consider their deployments at immediate risk. Any instance with port 8080 accessible from untrusted networks—whether directly exposed to the internet or reachable from internal networks where attackers may have a foothold—is vulnerable to exploitation.

FastGPT has released version 4.14.13 to address the vulnerability. Security teams should prioritize identifying all FastGPT deployments, checking version numbers, and applying the patch immediately. For instances that cannot be upgraded immediately, restricting access to port 8080 through network segmentation or firewall rules can serve as a temporary mitigation. Given the severity of unauthenticated RCE, organizations should also review logs for suspicious activity targeting the agent-sandbox service and consider the possibility of prior exploitation. The disclosure underscores the importance of securing AI infrastructure components—particularly those designed to execute arbitrary code—with proper authentication and network controls.