JupyterHub EUVD-2026-96917

Requires a multi-user JupyterHub instance (through 6.0.1) with the named-server feature enabled and with the attacker able to obtain or register an authenticated account whose username contains a hyphen that collides with the 'user-servername' OAuth client identifier of a specific victim (e.g. attacker 'alice-prod' to target user 'alice' server 'prod'). Exploitation is limited by: it requires an authenticated low-privilege account (PR:L), control over the attacker's own username to craft the collision, the named-server/OAuth-client feature being in use, and a specific pre-identified target - making this a targeted account-level denial rather than untargeted or default-configuration compromise.

This is a genuinely low-priority issue, and the signals are internally consistent rather than conflicting. The vendor-supplied CVSS 4.0 base score is 2.3 (Low) with vector AV:N/AC:L/AT:P/PR:L/UI:N/VC:N/VI:L/VA:L - network-reachable but requiring an authenticated low-privilege account (PR:L) plus a specific attack requirement (AT:P): the attacker must control a username whose hyphenated form collides with a victim's named-server OAuth client identifier. Impact is bounded to low integrity and low availability (VI:L/VA:L) with explicitly no confidentiality impact (VC:N) and no scope change to other systems (SC/SI/SA:N). No EPSS score, CISA KEV listing, or distinct proof-of-concept exploit flag was provided in the input; there is no public exploit identified at time of analysis, though a third-party technical write-up (hackmd.io) and a VulnCheck advisory describe the mechanism. The only notable tension is the 'Information Disclosure' tag, which contradicts VC:N - nothing is disclosed; the flaw lets an attacker overwrite another user's OAuth client to break that user's login and revoke their tokens, which is an integrity/availability denial-of-service against a specific targeted account, not data exposure. Net: a real but narrow, targeted abuse requiring a privileged precondition (an authenticated account plus a colliding username on a multi-user instance with named servers), not a mass-exploitable critical bug.

On a multi-user JupyterHub instance where users can self-select usernames and named servers are enabled, an authenticated attacker registers a username such as 'alice-prod' that collides with the OAuth client identifier generated for user 'alice' running a named server called 'prod'. By starting and then stopping their own named server, the attacker overwrites and then invalidates alice's OAuth client, breaking alice's OAuth login flow and revoking her active tokens. No public exploit code is identified, but a detailed technical write-up describing the collision exists.