Maximo AI Service CVE-2026-90920

The description attributes the flaw to 'improper control of code generation' (CWE-94) in IBM Maximo AI Service, but does NOT name a specific feature toggle, protocol, or non-default configuration as a prerequisite. Per the CVSS vector (AV:N/AC:L/PR:N/UI:N), no special conditions are required beyond network reachability: remote unauthenticated exploitation is possible against reachable instances of Maximo AI Service 9.1.0 and 9.2.0. The primary real-world LIMITING factors are therefore exposure-based rather than configuration-based - the vulnerable instance must be network-reachable by the attacker, and only versions 9.1.0 and 9.2.0 are in scope (patched builds are not). Note the description does not explicitly confirm whether the affected code-generation path is reachable in all default deployments; defenders should verify endpoint exposure and version with IBM advisory 7290833.

Multiple signals align toward genuine high priority. The CVSS 3.1 vector (AV:N/AC:L/PR:N/UI:N) describes unauthenticated, low-complexity remote exploitation with no user interaction and full CIA impact (C:H/I:H/A:H), consistent with the CWE-94 'improper control of code generation' (code injection) root cause and the stated outcome of arbitrary code execution. A vendor patch is confirmed available, which both validates the vulnerability's reality and indicates IBM treated it as fixable/serious. However, several enrichment signals are MISSING from the input: there is no EPSS exploitation-probability score, no CISA KEV listing, no SSVC decision data, and no indication of public POC code - so real-world exploitation likelihood cannot be quantified. The absence of KEV and POC does NOT mean low risk; it means exploitation activity is simply unmeasured here. Net: this is a credible critical-severity RCE where the technical risk is high and the exploitation-in-the-wild evidence is currently absent/unknown rather than contradictory. No signals conflict with one another; the limitation is incompleteness, not disagreement.

A remote, unauthenticated attacker sends a crafted request to an exposed IBM Maximo AI Service instance (9.1.0 or 9.2.0) that reaches the vulnerable code-generation path, injecting attacker-controlled code that the service executes. Successful exploitation yields arbitrary code execution in the service context, enabling full compromise of the host and the enterprise-asset data it processes. No public exploit code was identified in the provided data, so this scenario is inferred from the CVSS vector and CWE-94 classification rather than from a demonstrated POC.