Week of August 24, 2026: A Plain-English Roundup For Busy Leaders.

Microsoft has published firm dates for ending Microsoft-provided SMS and voice multifactor authentication in Entra ID, the service behind your Microsoft 365 logins. Starting September 1, 2026, tenants with users set up for text or phone-call codes will have those users automatically enabled for passkeys, and Microsoft will begin nudging them to register one at their next sign-in. Users can snooze the prompt.

The hard date is February 1, 2027. After that, Microsoft-provided SMS and voice are retired. Anyone whose only sign-in method is a text or a phone call will hit a registration prompt they can’t dismiss. There is no opt-out from that enforcement — it applies to every tenant. Organizations with a genuine regulatory need for phone-based codes can bring their own telecom provider, with configuration available from October 30, 2026.

Sources: Microsoft Learn — Microsoft-provided SMS and voice retirement

On August 18, 2026, Microsoft merged the Microsoft 365 Copilot app and the consumer Copilot app into a single app called Microsoft Copilot — new name, new icon, and a new commercial web address moving from m365.cloud.microsoft to copilot.cloud.microsoft, with an automatic redirect. Your security, compliance and privacy settings didn’t change, and because the underlying application ID is the same, existing policies keep working. Early desktop preview opened August 18, with broader rollout beginning mid-September.

Three features were retired the same day: Copilot Group Chat, Podcasts and Deep Research in the consumer app. Content in Group Chat threads was not carried forward. The merged app does add clearer visual separation between work and personal accounts — account labels, a green shield for work accounts and different background colors.

Sources: Microsoft Support — Changes to the Microsoft Copilot app · Microsoft Partner Center announcements, August 2026

Microsoft is finishing its move to OneDrive-backed whiteboards and retiring the older ones. The migration tool for legacy enterprise whiteboards was retired on August 22, 2026 — that date has already passed. On September 5, 2026, remaining legacy whiteboards are permanently deleted, with no recovery option. The standalone Whiteboard app is deprecated on September 14, 2026, after which enterprise users work with Whiteboard inside Teams.

Sources: Microsoft 365 message center — MC1441775, Whiteboard retirement

In a message center post published around August 21, 2026, Microsoft announced that the usage reports in the Microsoft 365 admin center are moving to new web addresses as part of a domain consolidation. Nothing about the reports, the data or who can see them changes. Old and new addresses run side by side for at least 30 days, and any redirect happens no earlier than September 17, 2026. Commercial tenants move first, with government clouds following through early September.

—

Sources: Microsoft 365 message center MC1459136 — usage reports moving to new domains

On August 18, 2026, Microsoft moved multi-tenant agent management into public preview in the Microsoft 365 admin center. Administrators can now see a consolidated list of AI agents across the tenants they manage, add agents, install them across selected tenants, review their permissions and — this is the important verb — block them. Cross-tenant access is authorized through delegated admin permissions, so nobody gets standing access they shouldn’t have. Managing agents needs no extra license; seeing agent risk and activity requires a Microsoft Agent 365 license.

Sources: Microsoft 365 message center MC1456781 — public preview, manage agents across multiple tenants · Multi-tenant agent management in the Microsoft 365 admin center · Microsoft Learn — agent settings in the Microsoft 365 admin center

Researchers at Varonis disclosed 3 chained weaknesses in Copilot Personal — the consumer assistant at copilot.microsoft.com, not Microsoft 365 Copilot. Nicknamed CoSnitch and tracked as CVE-2026-24301, the flaw meant a specially built link could make Copilot run an attacker’s instructions the moment the page loaded, inside the victim’s own signed-in session. In testing, the researchers pulled email bodies and sender details, calendar entries with attendees, file names from cloud storage and prior chat history — all from accounts the user had already connected. Microsoft shipped the fix on August 18, 2026, and the researchers found no evidence anyone had used it in the wild.

There’s a second, more stubborn piece. A malicious web page, once summarized by Copilot, could write instructions into the assistant’s long-term memory — where they survived password changes, session revocation and device re-enrollment, and stayed active until the user deleted them by hand.

Sources: Varonis Threat Labs — CoSnitch · Microsoft Security Update Guide — CVE-2026-24301 · The Hacker News coverage

Two notable posts from OpenAI in one week. On August 18, 2026, the company disclosed that it had temporarily paused reinforcement-learning training on its newest models for about 2 weeks while it hardened and tested its own research environments, and that its largest planned frontier training run remains on hold. The reasons were a security incident involving a partner and early evidence that an upcoming model may reach the “critical” cybersecurity threshold in OpenAI’s own safety framework. The new controls it described are unglamorous and familiar: sandboxing, network isolation, continuous testing and monitoring — at roughly 20% additional compute cost.

Then on August 19, OpenAI committed to keeping zero data retention available for eligible API customers and previewed Private Safety Processing, an approach where safety monitoring happens on customer-controlled infrastructure, or on OpenAI infrastructure encrypted with keys the customer holds. OpenAI receives only an alert category and severity, never the content. It’s in testing with early customers, with broader rollout planned for September.

—

Sources: OpenAI — Pacing model development in an era of cyber-critical capabilities · OpenAI — Offering zero data retention for frontier models

CVE-2026-68820 is a Windows flaw in a core networking driver that lets someone who already has a low-privilege foothold on a machine promote themselves to full system control. Microsoft patched it on August 11, 2026, and CISA added it to the Known Exploited Vulnerabilities catalog the same day, because it was already being used in real attacks before the patch existed. Under CISA’s risk-based timelines, internet-facing Windows systems were due by August 14 and internal Windows endpoints are due August 25, 2026 — tomorrow. There’s no workaround; the driver can’t be disabled or firewalled.

One detail matters more than the CVE number: the fix replaces a kernel driver, so the vulnerable version stays loaded until the machine restarts. A patch report showing 100% installed and 0% rebooted is not a patched fleet. For scale, Microsoft’s August update covered roughly 400 vulnerabilities including 3 zero-days — and July was larger still, at around 570.

Sources: Qualys — CVE-2026-68820 and CISA BOD 26-04 requirements · BleepingComputer — August 2026 Patch Tuesday · Tenable — August 2026 Patch Tuesday analysis

Researchers at Hudson Rock reported on August 16, 2026 — picked up widely on August 17 and 18 — that a threat actor is advertising large employee directories that appear to have been exported directly from organizations’ Microsoft Entra tenants. The listed data includes names, work email addresses, phone numbers, employee IDs, job titles, departments, reporting lines, group memberships, service accounts and lists of who holds global administrator rights. Named organizations include several very large global brands.

The researchers were careful about the cause, and so are we: they said the method isn’t conclusive, but assessed it’s most likely that attackers used credentials harvested by information-stealing malware on employee devices, not a flaw in Microsoft’s cloud. Related work from Arctic Wolf in early August described a parallel pattern — voicemail-themed phishing that captures both the password and the multifactor code in real time, after which attackers quietly read payroll, invoice and banking email without changing a single setting.

Sources: Hudson Rock — Azure exfiltration campaign · Arctic Wolf — Payroll Pirates: strange new tides in business email compromise

On August 18, 2026, CISA, the FBI and the Department of Health and Human Services published an update to their joint Medusa ransomware advisory. The victim count rose from over 300 to over 500 organizations across critical infrastructure, based on FBI investigative findings through April 2026. HHS was added as a co-author, reflecting how heavily healthcare has been targeted. The advisory also expanded its list of software flaws the group exploits to get in.

The two entry points haven’t changed: phishing for credentials, and unpatched software exposed to the internet. What’s worth reading closely is what happens next. Medusa operators abuse legitimate remote-support tools — the advisory names AnyDesk, Atera, ConnectWise, N-able, SimpleHelp, Splashtop and others — because those tools look like normal IT activity. Then they delete backup snapshots before encrypting.

—

Sources: CISA — #StopRansomware: Medusa Ransomware (AA25-071A) · Joint advisory PDF, updated August 18, 2026