Authlib CVE-2026-104056

Requires that the attacker be able to poison or intercept the HTTP response returned when Authlib (version ≤1.7.2) fetches server discovery/metadata JSON - i.e. an attacker-in-the-middle position on the metadata fetch, a compromised network path, or a malicious/compromised discovery endpoint the client is pointed at. The flaw is that Authlib caches these discovery values without validating or origin-binding the endpoint URLs to the configured issuer/metadata URL, so substituted endpoints are trusted. Limiting factors: a privileged network position or control of the discovery response is required; metadata fetched over properly validated TLS directly from a trusted issuer is not exploitable without first defeating that channel, which meaningfully constrains realistic exploitation despite the critical impact.

Signals conflict sharply and should be weighed carefully. The input CVSS of 9.8 (AV:N/AC:L/PR:N/UI:N, C:H/I:H/A:H) and SSVC (Technical Impact: total, Automatable: yes) portray a trivially-exploitable critical flaw, and CWE-345 (insufficient verification of data authenticity) matches the described failure to bind discovery-endpoint URLs to the issuer origin. However, EPSS is only 0.10% (1st percentile), the CVE is NOT in CISA KEV, and SSVC exploitation is merely 'poc' - so there is no evidence of real-world or widespread exploitation. More importantly, the description's exploitation primitive is a 'poisoned discovery response,' which in practice requires an attacker-in-the-middle position on the metadata fetch or a compromised/malicious discovery endpoint; this is inconsistent with the input vector's AC:L/no-MITM assumption (I independently assess AC:H - see assessed vector). Net judgment: this is a genuine high-impact design flaw in a trust-critical OAuth/OIDC code path and should be patched promptly, but it is NOT a drop-everything mass-exploitation priority - the very low EPSS and the privileged-network-position prerequisite point to targeted, not opportunistic, risk.

An application uses Authlib (≤1.7.2) as an OAuth/OIDC client and fetches its provider's discovery metadata over the network. An attacker able to intercept or poison that discovery response substitutes the authorization/token/userinfo endpoints with attacker-controlled URLs; because Authlib caches the metadata without binding endpoints to the issuer origin, the client then sends credentials, authorization codes, or tokens to the attacker. A proof-of-concept is referenced (SSVC: poc) via a public write-up, though no CISA KEV active-exploitation listing exists.