SOC Weekly Brief The week in the Microsoft security stack, distilled

Latest issue · Week 29 · 6 min read

July 13 – July 20, 2026

Act by

  • 14 Jul 2026 — July's Patch Tuesday is the largest on record, fixing roughly 570 vulnerabilities including 59 critical and three zero-days. Two are already being exploited: CVE-2026-56155, an Active Directory Federation Services (AD FS) elevation-of-privilege flaw, and CVE-2026-56164, a missing-authentication elevation-of-privilege bug in SharePoint Server; the third, CVE-2026-50661, is a publicly disclosed BitLocker bypass requiring physical access. Prioritize the AD FS and SharePoint fixes on identity-federation and internet-facing systems now, and hunt for exploitation on those hosts rather than waiting for the full-fleet rollout. (MSRC Security Update Guide)

  • 1 Aug 2026 — The Microsoft Azure Network Adapter (MANA) rollout reaches MANA-eligible Intel v1–v4 VM sizes in the public cloud, the earliest date those VMs can be placed on MANA-capable hardware. Network Virtual Appliances — including third-party firewalls and other security appliances — depend directly on the underlying network hardware and drivers, so an NVA whose OS or vendor build doesn't support MANA can see degraded or interrupted traffic. If yours isn't confirmed MANA-compatible, apply and enable the LegacyVMNVA Azure Policy tag before this date to hold those VMs off MANA hardware (honored until 31 May 2027); the MANA-eligible Cobalt 100 and Intel v5 sizes already passed their 26 May 2026 date. (Microsoft Learn)

  • 1 Sep 2026 — Passkeys become the default authentication method in Microsoft Entra ID. From this date, users enabled for SMS or voice are automatically enabled and nudged to register a passkey at multifactor sign-in, and native SMS and voice authentication is then discontinued on 1 Feb 2027, after which affected users must register a passkey to sign in with no opt-out. Identify who still relies on telephony MFA, enable synced or device-bound passkeys, and run registration campaigns now rather than at cutover; monitor for bypass attempts during the transition. (Microsoft Security Blog)

What changed

Microsoft launched Microsoft Defender Experts Threat Intelligence and, in the same post, renamed Defender Experts for XDR to Microsoft Defender Experts MDR. The new threat-intelligence service delivers expert-curated insight tuned to a customer's industry, geography, and environment, with early-warning alerts on emerging campaigns and recurring analyst briefings. Defender Experts MDR keeps its Microsoft-focused coverage as Plan 1 and adds a Plan 2 that extends 24/7 managed monitoring to third-party and multicloud sources through Microsoft Sentinel integration. For a SOC weighing managed detection, the split makes the multi-vendor coverage boundary explicit and pulls non-Microsoft signal into the same expert-run investigation loop. (Microsoft Security Blog)

Microsoft Sentinel announced an agentic experience for building ASIM parsers, released as open source. Creating a production-ready parser has always meant deep knowledge of the source data and the ASIM schemas, careful KQL design, and repeated validation cycles; the new workflow packages that as a set of skills the agent walks through — requirements, parser generation, then validation — rather than a single hopeful prompt. Analysts invoke it by chatting with GitHub Copilot in the Microsoft Sentinel Visual Studio Code extension, which loads the parser-creation skills. Because ASIM parsers are the normalization layer that makes detections, hunting, and analytics portable across sources, faster parser creation shortens the path from a new log source to working content. (Microsoft Sentinel Blog)

Microsoft published field notes from an Identiverse 2026 roundtable of 150 identity professionals that read as a governance checklist for AI agents. Nine in ten tables described agent sprawl as a present reality rather than a future risk — one organization discovered 44,000 agents in its tenant by accident — and the recommended controls are concrete: every Microsoft Entra Agent ID identity carries a sponsor (accountable for what the agent does) and an owner (responsible for its technical management), and Entra lifecycle workflows can automatically trigger an ownership review for every agent tied to a person when that person leaves. Microsoft Agent 365 sits on top of Entra Agent ID as the single control plane for seeing and acting on every agent. For a SOC, an orphaned or over-scoped agent surfaces through the same ownership and lifecycle machinery that already governs human accounts instead of lingering as an unmanaged credential. (Microsoft Entra Blog)

Microsoft Defender for Cloud took its CSPM assessments for Azure Database for PostgreSQL Flexible Server to general availability. The built-in assessments continuously evaluate PostgreSQL server configuration against PostgreSQL-specific security best practices — spanning network security, auditing controls, and operational resilience — and surface findings directly in Defender for Cloud so misconfigurations can be prioritized by risk. They light up automatically on Flexible Server instances where Defender for Cloud is already enabled, with no extra setup or cost, and more assessment coverage is planned. For a SOC this closes a posture-visibility gap on a common managed-database tier without standing up anything new. (Microsoft Community Hub)

Worth knowing

Microsoft Threat Intelligence detailed a compromise of the @asyncapi npm organization, where a malicious payload was inserted into one source file per package across five versions of four packages and republished within about 90 minutes. Unlike the usual postinstall trick, the payload runs at import time — so it slips past npm install --ignore-scripts — spawning a hidden detached process that pulls an ~8.2 MB encrypted bundle from IPFS and unpacks the Miasma command-and-control runtime through three static-key decryption layers. The root cause was a misconfigured GitHub Actions workflow using pull_request_target that leaked a bot personal-access token, letting the attacker publish through legitimate trusted-publishing channels with valid OIDC provenance. Defender Antivirus flags the artifacts as Trojan:JS/MiasmStealer.SC and Trojan:Script/Supychain.A; remove the affected versions, purge npm and Yarn caches, hunt for a sync.js under Node.js masquerade paths, block outbound traffic to 85.137.53[.]71 on ports 8080/8081/8091, and rotate credentials from any affected build system. (Microsoft Security Blog)

Microsoft also dissected two active intrusion chains delivering ACR Stealer, an information stealer sold as a service and tied to a rebrand of Amatera Stealer that aggressively harvests browser credential stores, session tokens, and sensitive documents. Both chains open with ClickFix paste-and-run social engineering; one then uses WebDAV-mounted shares and Python loaders with blockchain-backed dead-drop C2 resolution, while the other runs fileless through MSHTA and hides its payload in image pixels via steganography, executing entirely in memory. Defender's coverage is behavioral — living-off-the-land execution, suspicious WebDAV and MSHTA activity, obfuscated PowerShell, scheduled-task persistence, and DPAPI credential access — so the highest-leverage control is blocking ClickFix-style prompts and constraining PowerShell and Python from untrusted sources. Reported C2 includes looksta[.]icu, apigrokcloud[.]icu, and enhanceblabber[.]cc. (Microsoft Security Blog)

Microsoft published a least-privilege pattern for AI agents that is worth a read before agent identities pile up in your tenant. It argues for treating each agent as a first-class principal with a dedicated, named-owner identity rather than a shared service account, then layering task-based RBAC, resource-and-sensitivity scoping, tool binding through curated allowlists, just-in-time elevation for temporary privilege, and end-to-end auditing that captures agent identity, role, scope, resource, action, and timestamp. The security relevance is that agents chain individually low-risk actions across systems into paths that enable data exfiltration or privilege escalation, and the audit trail is what lets an analyst reconstruct that path after the fact. Microsoft frames the enforcement around Entra for identity and access and Purview for data governance. (Microsoft Security Blog)

Ahead of Black Hat USA 2026, Microsoft previewed its themes for the week — defending against AI-enabled threats and supply-chain attacks, and helping teams find the trust paths adversaries exploit — and pointed at the same Defender Experts threat-intelligence and multicloud MDR work rather than announcing new platform features. It is a useful signal of where Microsoft is steering its security messaging, but nothing in it needs action on shift. (Microsoft Security Blog)