Two 2026 CVEs in our source data carry serious severity signals — a VulnCheck KEV listing for Milvus, the vector database widely used for RAG and embedding search in AI applications, and a maximum CVSS 10.0 for Azure AI Language — but almost no technical detail beyond that in our sources. We’re reporting what we actually have rather than filling the gap with plausible-sounding specifics; the thinness of the evidence is itself the most important fact for deciding how urgently to act on either one.
CVE-2026-26190: Milvus, missing authentication for a critical function, actively exploited
Affects Milvus (vendor and product both recorded simply as “milvus” in our source). The only source in our ledger is VulnCheck’s KEV catalog, which lists this CVE as under active exploitation (added 2026-08-13) and titles it “milvus milvus Missing Authentication for Critical Function.” VulnCheck’s mitigation guidance is generic: apply the vendor’s remediation or discontinue use of the product if none is available. FIRST EPSS scores this CVE at 36.9% (98th percentile) — a materially elevated predicted-exploitation likelihood, consistent with the KEV listing. Our ledger has no CVE description, no CWE assignment beyond what the title implies, no CVSS score, and no publication date. Confidence: medium on the bare fact of the vulnerability and its active-exploitation status — VulnCheck KEV is a single but authoritative source for that specific claim — with no independent technical detail available to assess beyond it.
CVE-2026-70352: Azure AI Language, maximum-severity elevation of privilege, minimal detail
Affects Azure AI Language (Microsoft). The only source in our ledger is Microsoft’s own MSRC record, which assigns CVSS base 10.0 (critical; vector AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H) — the maximum possible score — and CWE-306 (Missing Authentication for Critical Function), titled “Azure AI Language Elevation of Privilege Vulnerability.” Our ledger has no description text beyond that title, no publication date, and no KEV listing in any catalog. Confidence: medium — MSRC only, and the specific mechanism behind the privilege elevation isn’t in our data. Given the maximum CVSS score, MSRC’s own advisory page should carry substantially more operational detail than has reached our ledger; treat that advisory as the authoritative next step rather than this summary.
Confidence and evidence gaps
Both CVEs trace to exactly one authoritative source each — VulnCheck KEV for Milvus, MSRC for Azure AI Language — which is enough to trust the severity signal (confirmed active exploitation in one case, a maximum CVSS score in the other) without being enough to describe the underlying mechanism, affected versions, or remediation steps. That’s a real gap in our current source coverage for both, not an editorial choice to withhold detail.
Why this matters
Vector databases like Milvus increasingly sit underneath RAG pipelines and hold embeddings derived from potentially sensitive source documents, so an unauthenticated-access flaw with confirmed active exploitation deserves prompt attention even without a full technical writeup. A maximum-severity CVSS score on a managed Azure AI service is similarly not something to wait on for more detail to accumulate — both cases call for checking the vendor’s own advisory directly and patching on that basis, rather than waiting for this pipeline’s source data to fill in.
Frequently Asked Questions
Why is there so little detail on these two CVEs? Because that’s genuinely all our current source data contains from the sources we ingest for each — VulnCheck KEV for Milvus, MSRC for Azure AI Language. We’re reporting the gap rather than guessing at specifics.
Is Milvus being actively exploited? Yes — VulnCheck’s KEV catalog lists CVE-2026-26190 as under active exploitation.
Is Azure AI Language being actively exploited? No KEV listing appears in our data for CVE-2026-70352, despite its maximum CVSS score.
What should I do given the limited detail? Check Milvus’s own security advisories and the Microsoft Security Response Center portal directly for these CVE IDs, and patch on that basis rather than waiting for more detail to reach this pipeline.
Data sourced from Microsoft Security Response Center (MSRC), VulnCheck KEV, and FIRST EPSS records, evaluated September 2026. See more vulnerability intelligence.