An OS command injection flaw in the shell-based temporary-file cleanup routine of Progress Software's Autonomous REST Connector GenAI Agents ARCGenAI-Generator version 2.0 allows an attacker to execute arbitrary shell commands on the machine that runs the generator. The vulnerability arises when the generator processes a Swagger/OpenAPI document supplied by the attacker; if the document is crafted, the cleanup logic treats it as input and spawns a shell, giving the attacker full control of the local environment. This leads to a remote code execution (RCE) scenario that can compromise confidentiality, integrity, and availability of the developer’s system. The weakness is classified as CWE-78, which indicates OS command injection.

The affected product is Progress Software's Autonomous REST Connector GenAI Agents, specifically the ARCGenAI-Generator component delivering AI-powered model generation from Swagger/OpenAPI definitions. Version 2.0 is vulnerable; all later releases, starting with 2.1, are presumed fixed. No other versions were noted as affected.

With a CVSS score of 9.6 the risk is considered critical. An attacker does not need direct network access; they only need to supply a malicious Swagger/OpenAPI file that is subsequently processed by a user running the generator. The exploit requires the victim to invoke the generator on the tainted document, so it can be triggered by a compromised collaboration platform or a social-engineering attack that leads the developer to publish the file. Because the vulnerability depends on local execution, mitigations are limited to preventing the processing of untrusted documents or updating to a patched version. The vulnerability is not listed in CISA's KEV catalog, and the EPSS score is 2%.