datamodel-code-generator CVE-2026-55403

The operator must specify authentication credentials via --http-headers or --url userinfo. The schema host must respond with an HTTP redirect (3xx status) to a different origin (different scheme, host, or port). An attacker could also inject such a redirect via a malicious $ref in the schema if the operator fetches a schema from an untrusted source. No authentication is required against the tool itself, but the attack complexity is high because it depends on the redirect behavior of the schema host. Authentication requirements: unauthenticated to the tool, but credentials must be supplied for the schema fetch.

CVSS 3.1 base score 3.7 (Low) with vector AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:N/A:N indicates a low-risk vulnerability: network-based but requiring high attack complexity (a cross-origin redirect) and only compromising low confidentiality (credential leak). No EPSS score is available, but given the specific preconditions and limited product footprint, the exploitation probability is likely very low. The CISA KEV does not list this CVE. There is a public PoC, but active exploitation is not confirmed. The SSVC decision would likely rate this as 'Unlikely' exploitation. The CVSS score accurately reflects the limited real-world risk, as successful exploitation depends on operators using authentication headers with a schema host that performs a cross-origin redirect.

An operator uses datamodel-code-generator with --http-headers 'Authorization: Bearer <token>' to fetch a schema from a trusted API. That API (or an attacker who can control the schema's $ref) responds with a 302 redirect to an attacker-controlled server. The tool resends the Authorization header to that server, leaking the token. A public PoC demonstrates this with two loopback servers.