A critical path traversal flaw in DB-GPT v0.8.1 and earlier lets unauthenticated remote attackers write arbitrary files anywhere on the server via a poisoned user_id HTTP header; a primitive that, depending on write target and host configuration, could plausibly be escalated to remote code execution.
What Is It
CVE-2026-73034 is an unauthenticated path traversal vulnerability (CWE-22) in eosphoros-ai's DB-GPT. The Python file-upload endpoint trusts the user_id HTTP header when building the destination path for uploaded content. An attacker sends a crafted multipart upload request with directory traversal sequences injected into that header, escaping the intended upload directory and writing attacker-controlled content to any location on the filesystem.
Per the NVD record, write targets include Python startup hooks, cron directories, and agent scripts. Writing to locations of that kind is the conventional route from arbitrary file write to code execution, though the record establishes the file-write primitive itself rather than a demonstrated end-to-end RCE chain; whether execution follows depends on the deployment's file permissions, process user, and whether those paths are actually loaded at runtime. VulnCheck ([email protected]) is the disclosing source.
Why It Matters
The vulnerability carries a CVSS 3.1 base score of 9.8 (CRITICAL) with vector CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H, network-reachable, low attack complexity, no privileges, and no user interaction required, with high impact to confidentiality, integrity, and availability. A CVSS 4.0 secondary score of 9.3 (CRITICAL) is also assigned.
Arbitrary file write on a network-exposed endpoint with no authentication is about as low a bar as exploitation gets, and it is a serious integrity compromise on its own terms even where code execution does not immediately follow. Because the traversal payload rides in an HTTP header rather than a filename, naive upload-path sanitization is unlikely to catch it.
CVE-2026-73034 does not appear in CISA's Known Exploited Vulnerabilities Catalog as of 2026-08-11, so active exploitation is not confirmed at this time.
What's Vulnerable
- Vendor: eosphoros-ai
- Product: DB-GPT
- Affected versions: all versions up to and including 0.8.1 (default status: affected)
No CPE entries were published with the record.
Patch Status
A fix commit exists upstream: e0c741bd2b5e521b128cffb3f68982dde3f7b359 in the eosphoros-ai/DB-GPT repository, with tracking in issue #3104. No fixed release version is stated in the NVD record, and because the CVE is not in the KEV catalog, no federal required action or due date applies. Operators running DB-GPT 0.8.1 or earlier should update to a build containing that commit and restrict network exposure of the upload endpoint in the interim.
The CVE record is in Received status as of publication on 2026-08-11, so analysis may still change.
Sources
- NVD, CVE-2026-73034 (record published 2026-08-11)
- VulnCheck Advisory; https://www.vulncheck.com/advisories/db-gpt-path-traversal-arbitrary-file-write-via-user-id-header
- DB-GPT Repository; https://github.com/eosphoros-ai/DB-GPT
- Fix Commit; https://github.com/eosphoros-ai/DB-GPT/commit/e0c741bd2b5e521b128cffb3f68982dde3f7b359
- GitHub Issue #3104; https://github.com/eosphoros-ai/DB-GPT/issues/3104
- CISA Known Exploited Vulnerabilities Catalog; https://www.cisa.gov/known-exploited-vulnerabilities-catalog