GT Edge AI CVE-2025-63663

Exploitation targets the /api/v1/conversations/*/files API of the GT Edge AI Platform (Community edition per the CPE) on versions before v2.0.10, abusing missing or incorrect per-object access-control checks so one actor can read files belonging to other users (an IDOR/broken-object-level-authorization pattern). The attacker must be able to reach this API over the network and supply a valid conversation/file path or identifier. Limiting factors: scope is confined to reading uploaded files (no integrity or availability impact), the product is a niche AI platform rather than a widely deployed service, and although NVD scores PR:N, the per-conversation nature of the endpoint suggests a session or account may in practice be needed to reach it - this authentication requirement is not confirmed from available data and should be validated against the deployment. Fixed in v2.0.10; updating is the remediation implied by the version boundary.

The signals are moderate and broadly consistent. The NVD CVSS vector (AV:N/AC:L/PR:N/UI:N) describes a network-reachable, low-complexity flaw with no user interaction and, as scored, no required privileges, yielding a 7.5 driven purely by confidentiality loss (C:H/I:N/A:N) - appropriate for an information-disclosure/broken-access-control (CWE-284) issue that exposes other users' files but does not alter data or availability. Exploitation likelihood signals are modest: EPSS is 0.29% (21st percentile), and the vulnerability is NOT in CISA KEV, so there is no confirmed active exploitation. The vuln.today decision label is 'this_week', a near-term but non-emergency priority. One tension to verify: NVD scores PR:N (unauthenticated) while a broken-access-control flaw on a per-conversation files API (/api/v1/conversations/*/files) often presumes an authenticated session that improperly reaches another tenant's objects (classic BOLA/IDOR) - the 'unauthorized attackers' wording in the description does not disambiguate whether a valid account is needed, so the true privilege requirement should be confirmed with the vendor. Overall this is a real, actionable confidentiality issue worth patching promptly, but the low EPSS, absence of KEV, and limited product footprint place it below internet-facing RCE/KEV-class emergencies.

An attacker interacts with the GT Edge AI Platform's /api/v1/conversations/*/files endpoint and manipulates the conversation/file identifier to retrieve files uploaded by other users, bypassing the intended access-control checks. Given AV:N/AC:L/UI:N, this requires only crafted API requests over the network with no victim interaction. Two public research references (p80n-sec GitHub/gist) and a VulDB advisory exist, but the input does not confirm any of them as working exploit code.