Summary
CVE-2026-75337 is a critical path traversal vulnerability (CWE-22) in the static resource interface of Yu AI Code Mother, an AI-powered code generation and one-click deployment platform built on Spring Boot. The flaw allows unauthenticated, remote attackers to read arbitrary files on the server, including application configuration and credential files, through the /api/static/{deployKey}/ endpoint. It carries a CVSS v3.1 base score of 9.8 (Critical), reflecting network exploitability, no authentication or user interaction requirements, and high impact to confidentiality, integrity, and availability.
Technical details
- Root cause: The static resource handler builds the file path to serve by directly concatenating an attacker-controlled path segment onto the preview root directory (e.g.
PREVIEW_ROOT_DIR + "/" + deployKey + resourcePath) without normalizing or validating the resulting path. - Missing authentication: The vulnerable endpoint does not carry the application’s
@AuthCheckannotation used elsewhere in the codebase, so the AOP-based authentication layer is bypassed entirely — the route is reachable by anonymous, unauthenticated requests. - Trigger conditions: An attacker sends a GET request to
/api/static/{deployKey}/and appends directory traversal sequences (e.g.../../) to the resource path to escape the intended per-deployment preview directory ({app}/tmp/code_output/{deployKey}/) and reach files elsewhere on the filesystem, such as application temp directories containing configuration or secrets files. - Attack vector: Network-based, requiring only a crafted HTTP GET request against the exposed application — no credentials or victim interaction needed.
- Impact: Disclosure of sensitive files (e.g., database credentials and other secrets stored under application temp/config directories). Depending on deployment configuration (reverse proxies, containers) the reachable file scope may extend further, and disclosed credentials/config data could enable follow-on compromise of backend systems.
- Note on exploitation constraints: Testing against embedded Tomcat 10.1 showed that the container blocks requests containing three or more
../sequences and refuses certain URL-encoded traversal variants, but the underlying flaw remains exploitable and impact can be greater depending on the front-end/reverse proxy configuration.
Affected software
- Yu AI Code Mother (
yu-ai-code-app/ yu-ai-code-mother-microservice), version v4.3 - Deployed on Spring Boot 3.5.3 with embedded Tomcat 10.1 and JDK 21 (as observed in the analyzed build)
Severity
- CVSS v3.1 Base Score: 9.8 (Critical)
- Vector:
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
Mitigation and recommended actions
- Immediate: No official vendor-published patched version has been confirmed at the time of this writing. Security teams should monitor the Yu AI Code Mother project for an updated release that adds path normalization/canonicalization and enforces authentication on the
/api/static/{deployKey}/endpoint, and upgrade as soon as a fix is available. - If no patch is available:
- Restrict or disable public network exposure of the
/api/static/{deployKey}/endpoint until a fix is applied — place it behind authentication at the reverse proxy/gateway layer. - Configure a WAF or reverse proxy rule to block requests to this path containing traversal sequences (
../, encoded variants such as%2e%2e%2f, etc.). - Run the application with a least-privilege filesystem context so that even if traversal succeeds, sensitive files (credentials, secrets, configuration) are not readable by the application process.
- Avoid storing plaintext secrets (e.g., database passwords) in locations reachable from the application’s temp/output directories; use a secrets manager or environment-injected secrets instead.
- Monitor logs for requests to
/api/static/containing directory traversal patterns as an indicator of exploitation attempts.
- Restrict or disable public network exposure of the