OpenShift AI CVE-2026-5483

Exploitation requires reaching the specific vulnerable NodeJS endpoint exposed by the odh-dashboard component of Red Hat OpenShift AI; per the CVSS vector (PR:L) the attacker needs low-level authenticated access to the dashboard, with no user interaction (UI:N). Impact is meaningfully gated by the RBAC bound to the disclosed Service Account - the leaked token is only as powerful as that SA's permissions, so clusters with tightly scoped SAs limit downstream damage. Limiting factors: the affected component is only present where RHOAI 2.16 or 3.3 is deployed, the dashboard must be network-reachable by the attacker, and authentication to the dashboard is needed under the stated vector. One caveat for defenders: the intelligence tags include 'Authentication Bypass', which would conflict with PR:L - if an unauthenticated route exists, the authentication prerequisite no longer limits exploitation; verify with the Red Hat advisory.

The signals are mixed and worth reconciling. CVSS 3.1 is 9.9 (AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H): network-reachable, low complexity, only low privilege required, no user interaction, with total compromise across a scope boundary - a severe score. However EPSS is just 0.06% (20th percentile) and CISA's SSVC framework records Exploitation: none and Automatable: no, while still rating Technical Impact: total. So severity-if-exploited is genuinely high, but the likelihood/automatability signals are low, pointing to a high-impact but not currently mass-exploited issue. There is no KEV listing and no POC referenced, so this is no public exploit identified at time of analysis - not active exploitation. Note one tension to verify: the CVSS vector states PR:L (authenticated), yet the intelligence tags include 'Authentication Bypass'; defenders should confirm with Red Hat whether any unauthenticated path exists, as that would materially raise real-world risk. Overall this is a real priority to patch given total technical impact and a network vector, but the low EPSS/SSVC exploitation signals suggest opportunistic mass exploitation is not yet occurring.

A data scientist or other low-privileged user with a valid odh-dashboard login (or anyone who can reach the dashboard if an auth-bypass path is confirmed) sends a request to the vulnerable NodeJS endpoint and receives a Kubernetes Service Account token in the response. They present that token to the Kubernetes API server and act with the SA's RBAC permissions to read secrets, modify, or delete cluster resources. No public POC is currently referenced, and the CVSS AC:L/PR:L profile means the technical bar is low for an already-authenticated user.