Microsoft Fixes CVSS 10.0 Azure AI Foundry Auth Bypass Allowing Network Privilege Escalation
Microsoft silently mitigated a CVSS 10.0 authentication bypass in Azure AI Foundry plus four other critical cloud privilege-escalation flaws. No customer action was needed and no exploitation has been observed. The fixes expose recurring authorization weaknesses across newly deployed AI services.
The advisory lists four additional critical cloud CVEs including command injection in Microsoft 365 Copilot (CVE-2026-85885) and improper authorization in Azure Database for PostgreSQL (CVE-2026-85878). All were scored 9.6-9.9 and described as fully mitigated server-side. The credited researcher Rémy Marot reported the Foundry issue; Microsoft states no in-the-wild exploitation evidence exists.
Procurement records and prior job postings for Azure AI Foundry show accelerated rollout timelines after the 2025 rebrand from Microsoft Foundry. The cluster of AI-service authorization defects within one month points to shared identity and access control code paths that were not fully hardened before public availability.
Windows out-of-band updates addressed two local escalation flaws (CVE-2026-62721 and CVE-2026-85921) reaching SYSTEM and VTL1. These were shipped separately from the monthly rollup that fixed 974 total issues, two of which are under active exploitation via the BlueMoon kit.
The pattern indicates Microsoft is prioritizing rapid AI platform deployment while deferring full authentication boundary validation until post-release patching cycles.
MSRC: At least two additional Azure AI platform CVEs above 9.0 will appear in the next 90 days.
Sources (3)
- [1]Microsoft Security Response Center Advisory(https://msrc.microsoft.com/update-guide/vulnerability/CVE-2026-85889)
- [2]The Hacker News Report(https://thehackernews.com/2026/09/microsoft-patches-cvss-100-azure-ai.html)
- [3]Proofpoint BlueMoon Exploit Kit Analysis(https://www.proofpoint.com/us/threat-insight/post/bluemoon-exploit-kit)