July 27 – August 3, 2026

Act by

• 27 Oct 2026 — Microsoft Defender for Cloud stops enabling Foundational CSPM by default on new Azure subscriptions and moves it to an opt-in model. Existing Azure subscriptions keep their current configuration and AWS and GCP environments are unaffected, so this is a question of what a newly created subscription looks like on day one rather than a change to what you already have. Foundational CSPM stays free and can be turned on at any time, but a subscription created after the cutover starts with no posture management until someone enables it, which is exactly the gap that gets missed when a project team spins up a subscription outside the usual process. If your landing-zone automation or subscription-vending pipeline assumes posture coverage arrives for free, add an explicit enablement step before late October. (Microsoft Defender for Cloud release notes)

What changed

Microsoft published detection and response guidance for CVE-2026-54121, nicknamed Certighost, an authentication-bypass vulnerability in Active Directory Certificate Services that lets an attacker obtain certificates for arbitrary domain computer accounts including domain controllers. The flaw sits in an AD CS enrollment fallback called a chase: when the CA cannot resolve the requester it contacts a client-supplied domain controller named in the cdc attribute and folds the response into the issued certificate without verifying it is talking to a legitimate DC. A DC certificate is a path to full domain compromise through Kerberos ticket abuse and DCSync. Exploitation needs several conditions together — an AD-integrated Enterprise CA, enrollment through the machine template, a default ms-DS-MachineAccountQuota or an already-controlled machine account, SMB and LDAP reachability to the CA, and any non-privileged domain account — and it affects Windows Server 2012 through 2025 including Server Core. A public proof of concept landed on 24 July; Microsoft says it has not seen confirmed threat-actor exploitation and that observed activity looks like security testing. The July 14 security update is the fix and nothing else replaces it. For visibility, turn on the Certification Services audit subcategory and set the CA audit filter so events 4886 and 4887 are recorded, then watch for the Defender for Identity alert Potential Certighost (CVE-2026-54121) AD CS abuse, which fires on the anomalous issuance pattern itself. (Microsoft Defender XDR Blog)

Microsoft Threat Intelligence detailed CaptiveCrunch, a campaign by Storm-2945, a sub-cluster of Midnight Blizzard tied to Russia's SVR, that targets corporate travelers through captive portal networks at hotels, conferences, and shared venues worldwide. Compromised guest Wi-Fi manipulates DNS and HTTP traffic to route users through actor infrastructure, where a ClickFix-style fake OS or browser update convinces them to paste and run a payload. Two malware families follow: CornFlake, a Go-based Windows RAT with keylogging, clipboard and screen capture, browser credential theft that defeats Chrome App-Bound Encryption, and encrypted C2 over ECDH P-256; and ChocoShell, a PowerShell infostealer that bulk-collects browser cookies, saved passwords, Microsoft 365 SSO tokens, and Wi-Fi credentials while disabling AMSI and using several UAC bypasses. The campaign also runs device code phishing against Entra ID authentication flows, with lookalike domains such as ms365-device and m365-owa. Defender for Endpoint covers the ClickFix and credential-theft stages, Defender for Identity flags anomalous OAuth device code authentication, and Entra ID Protection catches the risky sign-ins. The two mitigations worth doing this week are blocking device code flow in Conditional Access wherever it is not needed, and telling travelling staff to avoid guest Wi-Fi in favour of a hotspot or eSIM. (Microsoft Security Blog)

The final phase of Microsoft Defender Threat Intelligence convergence went generally available on 1 August, and the standalone MDTI pages — Intel Profiles, Intel Explorer, and Intel Projects — retired the same day. Everything moves to the Threat Analytics tab in the Defender portal rather than the Threat Intelligence tab, with no migration action required. Two pieces matter operationally. Entity pages for IPs, domains, URLs, and files now carry a Threat Intelligence Insights tab combining reputation scores, attributed reports, infrastructure relationships, services, certificates, DNS and WHOIS records, trackers, and sandbox analysis in one place. And the intelligence that previously needed the Premium MDTI connector is now served by the free Microsoft Threat Intelligence connector in Sentinel, with MDTI APIs available by default to Defender XDR and Sentinel customers without a separate MDTI SKU. The API surface is unchanged, so no endpoint mapping or code changes are needed, but analysts who bookmarked the old Intel Explorer pages will need retraining on where the same data now lives. (Microsoft Defender XDR Blog)

Microsoft opened public preview of Project Perception on 3 August, an agentic security system built around three coordinated agent teams: red agents that probe for paths to compromise before an attacker finds them, blue agents that investigate and reason over context to decide what is meaningful risk, and green agents that take corrective action and harden the environment. Announced on 27 July alongside MAI-Cyber-1-Flash, Microsoft's first in-house cybersecurity model, the system starts inside Microsoft Defender and initially targets software vulnerability management, with Microsoft citing 96% on the CyberGym benchmark for the MAI-Cyber-1-Flash configuration at roughly half the cost of comparable setups. Microsoft describes it as a closed loop that continuously discovers, evaluates, and improves posture across identities, endpoints, applications, data, clouds, and AI systems, with expansion to more Microsoft Security products planned. Preview scope is narrow enough that this is a pilot rather than a rollout, but it is worth understanding now how autonomous remediation would interact with your change-control process before anyone enables green agents in production. (The Official Microsoft Blog)

Microsoft Sentinel added custom detections support in Sentinel repositories in public preview, completing the detections-as-code story across the portfolio. Custom detection rules can now be managed in a connected GitHub or Azure DevOps repo alongside analytics rules, playbooks, parsers, and workbooks, reviewed through pull requests, and deployed through existing CI/CD pipelines or the Microsoft Security Bicep extension. Until now custom detections were the piece that had to be clicked through the portal while everything around them was version-controlled, which made drift between environments hard to see and rollbacks manual. Two preview limits to plan around: custom frequency for Sentinel-only data is not supported yet, and neither are custom details. (Microsoft Sentinel Blog)

Sentinel data lake also gained table insights in public preview, a monitoring view on the tables page in the Defender portal that gives table-level observability without writing KQL or opening a workbook. It surfaces ingestion volume split by analytics versus data lake tier, day-over-week fluctuation against the same day last week, the top five tables by daily volume, last data received per table, estimated daily ingestion cost, and a signed volume anomaly against baseline. The last-data-received and anomaly signals are the ones that catch a connector that has quietly stopped sending, which is the failure mode that hides longest because nothing alerts on the absence of logs. The same July update added connectors in preview for GitHub Enterprise audit logs via Azure Blob Storage with near real-time latency, Agari Brand Protection and Phishing Defense, and Airlock Digital application control and execution events. (Microsoft Sentinel Blog)

Defender for Cloud completed the retirement of legacy grouped recommendations, also known as sub-assessments, on 31 July. The deprecated data is no longer reachable through the API, and the Azure portal and Azure Resource Graph may take a few days to catch up. Individual recommendations, announced back in May, are the replacement. If any of your automation, reporting, governance workflows, or ARG queries still read sub-assessment data, they will now return nothing rather than error loudly, which is the more dangerous failure — validate them, and migrate any remaining disable rules over to exemptions. (Microsoft Defender for Cloud release notes)

Worth knowing

Microsoft's July security roundup gathers the month's shipping items in one place, and a few are worth a second look if you skimmed them at the time: prompt injection protection in Defender for Office 365 in preview, which isolates emails carrying malicious AI instructions before delivery; unified Defender posture and runtime protection for cloud agents in Microsoft Agent 365 at GA; Defender-Entra interconnected experiences at GA, letting the SOC disable a compromised identity under its own RBAC; and Defender Experts Threat Intelligence at GA. (Microsoft Security Blog)

Black Hat USA 2026 runs 1–6 August at Mandalay Bay, and Microsoft's sessions cluster on Wednesday 5 August: a keynote from Agentic Security CVP David Weston on defending when offense is cheap, a main-stage session from Microsoft Threat Intelligence on hunting supply chain attacks, and a talk on turning threat intelligence into expert-led defense. Expect the usual post-conference wave of disclosure and research to land in next week's window. (Microsoft Sentinel Blog)