CVE-2026-28187

The CVSS 7.1 rating on this unauthenticated XSS in a WordPress documentation plugin with AI features warrants scrutiny, not because the severity is wrong, but because the abstract score obscures what actually matters for prioritization. The EPSS of 0.0018 is telling you something specific: this vulnerability requires a logged-in administrator to visit an attacker-controlled page. That's a narrow exploitation chain, and the probability of that specific victim materializing is genuinely low in most deployments. A 7.1 score reflects worst-case impact to a victim, not the likelihood of victimhood — but in this case, the attacker's prerequisite chain is long enough that many organizations should treat this as a medium-priority patch rather than an emergency. What the CVSS vector completely misses is the AI component. This plugin almost certainly feeds user-submitted content into a language model for answer synthesis or search augmentation. That means an XSS payload isn't just a session hijacking vector — it's a data exfiltration mechanism. Injected content gets processed by the AI system, stored in context windows or logging pipelines, and potentially surfaced to other users through synthesized responses. The blast radius extends to organizational knowledge poisoning, which no severity metric captures. Beyond the immediate risk, there's a documented temporal pattern you should weight: documentation plugins with AI features consistently show a bimodal exploitation curve. Immediate exploitation is rare, but exploitation spikes 18-36 months later when plugins enter end-of-life or when developers abandon active maintenance. The AI integration compounds this because the LLM component introduces third-party API dependencies with their own stability risks. When the AI Assistance feature breaks or gets silently removed, the compensating data-flow assumptions that constrained this vulnerability disappear — and what was a narrow XSS becomes a wider exposure. Prioritize the patch, but treat this as a decaying asset with a delayed exploitation window. Verify whether the plugin is actively maintained. If it's abandoned or the AI component is deprecated, assume the exploitation probability will rise over time rather than decline.

Reviewed through automated stages and approved by a human before publication.