GT Edge AI CVE-2025-63665

Exploitation requires the attacker to reach and submit input to the GT Edge AI Community Edition 'Prompt window' on an instance running a version before v2.0.12, delivering a crafted JSON payload that the application processes without safe handling (CWE-94). The CVSS vector asserts no authentication (PR:N) and no user interaction (UI:N), but the description does not state whether the prompt interface is exposed anonymously - authentication requirements are not confirmed from available data, and a required login or non-exposed/localhost-only deployment would materially limit reachability. Only the Community Edition is listed as affected per the CPE.

The signals are mixed between a high nominal severity and limited corroborating exploitation intelligence. The NVD CVSS 3.1 vector (AV:N/AC:L/PR:N/UI:N, C:H/I:H/A:H, score 9.8) describes an unauthenticated, network-reachable, low-complexity remote code execution with total impact - the theoretical worst case. However, the only intelligence beyond the MITRE-sourced description is two independent-researcher references (a GitHub gist and a 'Vulnerability-Research' writeup by p80n-sec), with no EPSS score, no CISA KEV listing, and no SSVC decision data supplied in the input. Absence of KEV and EPSS means there is no evidence of widespread or automated exploitation despite the critical score. A key tension to flag: the vulnerability is triggered by 'injecting a crafted JSON payload into the Prompt window,' which implies interaction with an application's prompt interface that in many AI products sits behind a login - the PR:N (unauthenticated) rating may be optimistic and should be verified against the product's actual deployment/auth model. This is best treated as a high-severity-but-unverified-reach issue: patch promptly if GT Edge AI Community Edition is deployed, but the real-world risk is bounded by the product's very small footprint rather than by mass exploitation.

An attacker able to reach the GT Edge AI Community Edition prompt interface submits a specially crafted JSON payload through the Prompt window; the application processes the injected input unsafely (CWE-94 code injection), causing attacker-controlled code to execute on the host. The referenced researcher writeup and gist may contain proof-of-concept details, though no exploit was explicitly flagged as public in the input. Given AV:N/AC:L, a single malformed request to an exposed instance is sufficient to achieve full compromise.