Db Gpt CVE-2026-4504

Remote unauthenticated attacker can exploit SQL injection in eosphoros-ai db-gpt versions up to 0.7.5 via the /api/v1/editor/ endpoint. No authentication, user interaction, or special configuration required. Default installations are vulnerable.

This vulnerability presents significant real-world risk based on multiple signals. The CVSS 3.1 score of 7.3 (High) reflects a network-accessible attack vector requiring no privileges or user interaction (AV:N/AC:L/PR:N/UI:N), with impacts to confidentiality, integrity, and availability all rated Low (C:L/I:L/A:L). The attack complexity is Low, making exploitation straightforward. Critically, a public proof-of-concept exploit is available on Huntr (https://huntr.com/bounties/e32fda74-ca83-431c-8de8-08274ba686c9), significantly lowering the barrier to exploitation. The vendor's non-responsiveness to early disclosure attempts raises concerns about patch availability and ongoing support. SQL injection attacks are well-understood and highly automatable, making this vulnerability a practical target for both opportunistic scanning and targeted attacks. The combination of public exploit code, network accessibility, and no authentication requirement elevates this beyond a theoretical concern to an immediate priority for organizations running affected versions.

An unauthenticated remote attacker discovers the publicly available proof-of-concept exploit on Huntr and adapts it to target an internet-exposed DB-GPT instance. By sending specially crafted HTTP requests to the /api/v1/editor/ endpoint with malicious SQL payloads, the attacker successfully bypasses authentication checks, extracts sensitive database records including user credentials and proprietary AI model configurations, and potentially modifies or deletes critical data. The low attack complexity and availability of working exploit code enable even moderately skilled attackers to successfully compromise vulnerable instances within minutes of discovery.