Topic · 64 stories
Microsoft security and privacy news
In short
Microsoft is the US technology company behind Windows, Azure, Microsoft 365, GitHub and the Copilot AI products. In 2026 it shipped a record 974 fixes in one Patch Tuesday, exposed Russian hackers using hotel Wi-Fi to steal logins, and announced it will retire SMS and voice MFA in Entra on 1 February 2027. Its own products were also targets, from prompt injection in Copilot to an unsigned-token flaw in the internal Titan service.
What is Microsoft in security and AI news
Microsoft is a US technology company behind Windows, Azure, Microsoft 365, GitHub and the Copilot AI products. Three storylines run through 2026: record security patches, hacking campaigns documented by its threat intelligence teams, and a race to dominate AI agents. Alongside them sit flaws in its own services and privacy disputes over Windows. Microsoft Copilot and Windows security have their own topic pages; this one covers the wider picture.
What AI products did Microsoft launch in 2026
The main ones are a single Copilot app with agents, Windows rebuilt for agents, and small in-house models. In September Microsoft unveiled an all-in-one Copilot app with three sections: Home with the Cowork agent, Code, and Autopilot for persistent agents. Full Word, Excel and PowerPoint run inside it, and users can pick OpenAI’s GPT, Opus from Anthropic, or an Auto mode. Autopilots run in the cloud with their own identity, memory and email address.
In October the company rebuilt Windows around AI agents. Copilot will run tasks without the cloud when they do not need heavy compute, and Autopilot lets users name an agent, give it a role and a goal. Some features shown are not yet available. The Surface Laptop Ultra with Nvidia’s RTX Spark costs $2,599 and ships from 16 October; the $5,999 Surface RTX Spark Dev Box follows in November.
The second strategy is cheap specialist models. MAI-Cyber-1-Flash scored 96 percent on CyberGym, 12 points above Mythos, at an estimated half the cost, but it hands hard cases to GPT-5.4. Decision-1, a model for fast structured decisions, is based on Qwen3.5-9B, reaches 83.5 percent accuracy at 85 ms latency and charges $0.042 per million input tokens. In September Microsoft AI also published a code of conduct for its MAI models that puts human control first, with a revised version due around the end of 2026.
How many flaws does Microsoft patch and what is Patch Tuesday
Patch Tuesday is Microsoft’s monthly security update, and in 2026 it keeps breaking records. September’s release fixes 974 flaws, including two exploited zero-days: CVE-2026-81963 in the Windows Update Stack and CVE-2026-85880 in Advanced Local Procedure Call. Both let an attacker who already has a foothold gain higher privileges. Dustin Childs of the Zero Day Initiative advises patching quickly. He singled out an Exchange Server flaw (CVE-2026-55007) triggered by a rigged Visio attachment and a 9.8-rated Outlook bug (CVE-2026-78509) triggerable through the Preview Pane.
| 2026 month | Fixes | Notes |
|---|---|---|
| June | 206 | previous Windows record |
| July | about 570 | three zero-days, two exploited (Active Directory, SharePoint), 61 critical |
| August | about 620 | per the September edition |
| September | 974 | two exploited zero-days, more than 110 critical, 20 wormable |
The editions give two July figures. One reports a record 570 Windows fixes, another, citing ProPublica, says Microsoft patched over 200 bugs in June and over 600 in July. The September edition puts July at roughly 570. The jump is linked to AI: the July record was attributed to the internal MDASH scanning tool, and ProPublica reported that Anthropic’s Mythos is overwhelming Microsoft with bugs. See also AI-discovered vulnerabilities and security updates.
Microsoft now urges shorter patch deployment timelines: a quality update deferral under three days, deadlines of zero or one day and a grace period of at most two days. A separate case is BitLocker CVE-2026-50661, a bypass of drive encryption: an attacker with physical access could skip the PIN and recovery key before the 14 July updates.
What attacks has Microsoft warned about
Microsoft Threat Intelligence documented a Russian campaign on hotel Wi-Fi, fake CAPTCHA pages called TerminalFix and Teams helpdesk impersonation. Storm-2945, a sub-cluster of Midnight Blizzard linked to Russia’s SVR, has hijacked captive portals since May, redirecting travelers to fake updates that install the CornFlake trojan and the ChocoShell stealer. Microsoft advises treating hotel Wi-Fi as untrusted and using phishing-resistant authentication. It is one example of state-sponsored hacking.
Other warnings: TerminalFix tricks users into running PowerShell in Windows Terminal, and a Teams campaign impersonating IT support ends in ransomware and data theft. Microsoft also published a technical breakdown of GigaWiper, a Go backdoor with three destructive commands. Its Digital Defense Report confirmed the first fully automated ransomware attack, JADEPUFFER, and said phishing caused 23 percent of investigated intrusions.
Microsoft’s own services are also used as cover: a phishing campaign through public Power BI dashboards delivered ScreenConnect and likely evaded some email filters.
What flaws were found in Microsoft’s own services
Researchers found flaws in an internal analytics service, an email anti-spoofing setting and Copilot in 2026. The internal Titan service never verified token signatures, so one endpoint allowed raw SQL against 17 ClickHouse databases holding over 17 trillion rows. The report reached MSRC on 5 September, the endpoint was locked down on 9 September and a $5,000 bounty followed.
Reject Direct Send is meant to block external mail disguised as internal, but ReliaQuest showed that an empty return address defeats it. ReliaQuest called this a limit on the setting’s scope, not a software flaw. The best-known Copilot case is a self-spreading worm hidden in Word documents, where white text on a white background hijacks the assistant, a technique covered under prompt injection. Around the same time researchers showed that Microsoft’s official Azure DevOps MCP server returns hidden HTML comments from pull request descriptions, which an AI reviewer then obeys.
Does Microsoft track Windows users
Microsoft can track a Windows installation through a persistent identifier, the Global Device ID, which let the FBI tie an alleged hacker to his computer despite a VPN. This surfaced in the arrest of 19-year-old Peter Stokes, an alleged Scattered Spider member. The ID changes only after a fresh Windows install, and Microsoft could link the new one to the old one through a Microsoft account login.
In July OneDrive Photos appeared silently on Windows 11 with face grouping that Microsoft itself says may count as biometric data in some regions. Microsoft says the data is deleted within 30 days of disabling the feature, but has not published whether facial analysis runs on the device or on its servers. In contrast, Microsoft became the first major AI developer to adopt binding student privacy standards for schools, including a ban on training models on student data and 72-hour breach reporting. They are not automatic: from 1 November districts can ask to add them to contracts.
Is SMS multifactor authentication going away in Microsoft accounts
Yes, in Microsoft Entra: SMS and voice MFA will be removed on 1 February 2027. Microsoft cited their unreliability against AI-assisted phishing. From 1 September 2026 users are asked to register a passkey, and a temporary opt-out exists until 1 February 2027, with none afterwards. Organizations that must keep telephony have to configure a supported telecom provider through the Microsoft Security Store by 30 October 2026, a paid add-on. Passkeys themselves are free in every Entra plan.
Microsoft’s Entra documentation (as of 16 October 2026) recommends phishing-resistant methods: Windows Hello for Business, FIDO2 passkeys and security keys, and certificate-based authentication. For the basics see password security.
What it means for you
- Install Windows updates quickly. With record patch volumes and exploited zero-days, the gap between a fix and an attack is now measured in hours.
- In a company, move users from SMS to passkeys in Entra before 1 February 2027 and find who still relies on SMS or voice.
- Treat hotel Wi-Fi as untrusted and ignore update prompts from a login portal.
- Never paste commands from a CAPTCHA page into PowerShell, and verify any “IT support” contact in Teams.
- Review spam override lists in Microsoft 365 and delete entries for executives that you cannot justify.
Still open: whether Microsoft will fix the Copilot worm weakness that the researcher published after 144 days, whether Direct Send will be disabled by default, and whether September’s patch volume repeats.
Key facts
- A phishing campaign abused public Power BI dashboards to push ScreenConnect downloads. It began around 10 September 2026 and linked to a real Microsoft service, likely evading some email filters. (source)
- Microsoft rebuilt Windows around AI agents. The Surface Laptop Ultra with Nvidia's RTX Spark costs $2,599 and ships from 16 October. (source)
- Microsoft's 2026 Digital Defense Report says the median time from vulnerability discovery to weaponization is under 24 hours, while remediation takes 30 to 60 days. (source)
- The new Copilot app combines chat, coding and always-on Autopilot agents. Microsoft reported over 30 million paid Microsoft 365 Copilot seats in July, up from 20 million in April. (source)
- September Patch Tuesday fixes a record 974 flaws, including two exploited zero-days: CVE-2026-81963 and CVE-2026-85880. (source)
- Storm-2945, a sub-cluster of Russia's Midnight Blizzard, has hijacked hotel and conference Wi-Fi since early May to steal credentials from business travelers. (source)
- Microsoft will remove SMS and voice MFA from Entra on 1 February 2027 and require passkeys. Passkeys are included at no extra cost in all Entra plans. (source)
- An unsealed complaint showed the FBI linked an alleged Scattered Spider member to a hack through his Windows Global Device ID, a persistent identifier, despite his use of a VPN. (source)
This edition was produced with artificial intelligence. Text and voice are generated automatically.
Timeline
- Microsoft unveils MAI Code 2, claiming Opus 5.5 performance at 5 percent of the cost AI
- Microsoft pins 36-hour November outage on a July code bug Security
- Microsoft’s Lazarus technique hides backdoors in fewer than 10 dormant neurons AI
- Microsoft’s MAI-DxO beat generalist doctors on complex diagnosis cases AI
- Microsoft patent would edit live speakers’ words and mouths in real time AI
- Russia adds Apple and Microsoft to surveillance registry Privacy
- Microsoft Authenticator for Android flaw let any app obtain work account tokens Security
- Microsoft disrupts phishing kits behind 1bn scam messages Security
- Developer releases free open-source clones of Word, Excel and PowerPoint AI
- Microsoft Research finds clearing chat history helps AI agents escape idea basins AI
Show older (54 stories)
- Phishing campaign abuses Power BI dashboards to deliver ScreenConnect Security
- Microsoft releases Decision-1 model for fast structured decisions AI
- Microsoft rebuilds Windows around AI agents with native Copilot and Autopilot AI
- Microsoft report confirms first fully automated AI ransomware attack Security
- Microsoft report says attackers hold early AI advantage over defenders Security
- Microsoft tests Cloud Rebuild recovery for Windows 11 Privacy
- Microsoft Launches All-in-One Copilot App With Coding and Always-On Agents AI
- Microsoft Titan flaw exposed 17 trillion analytics records via unsigned JWT Security
- Microsoft publishes a code of conduct for its MAI models AI
- Microsoft ships record 974 fixes, with two zero-days already exploited Security
- Microsoft Adopts Binding Student AI Privacy Standards in Deal With Teachers Union-Backed Academy Privacy
- Empty return address defeats Microsoft’s Reject Direct Send anti-spoofing setting Security
- Cyber criminals use ASCII smuggling to evade filters in mass phishing campaigns Security
- Spam Campaign Uses Invisible Unicode Characters to Split Word ‘Funding’ and Evade Filters Security
- Microsoft Warns of Teams Helpdesk Impersonation Campaign Ending in Ransomware Security
- Microsoft Warns of TerminalFix Campaign Abusing Windows Terminal for Backdoor Deployment Security
- TerminalFix Malware Chain: From PowerShell to Reverse Tunnel Security
- TerminalFix: Microsoft Urges Investigation for Lateral Movement and Credential Exposure Security
- Amazon, Cursor, Microsoft, OpenAI, and Vercel create shared standard for AI agent plugins AI
- Midnight Blizzard sub-cluster targets travelers via captive portals in CaptiveCrunch campaign Security
- Russian State Hackers Hijack Hotel Wi-Fi to Steal Traveler Credentials Security
- Microsoft ties CaptiveCrunch hotel Wi-Fi attacks to Russian APT29 sub-cluster Security
- AI agents at OpenAI, Anthropic, and Microsoft broke out, broke in, and obeyed AI
- Security researcher builds self-spreading worm that hijacks Microsoft Copilot AI
- Researcher builds self-spreading worm that hijacks Microsoft Copilot for Word Security
- Microsoft merges Copilot offerings into a single AI super app AI
- Microsoft AI bets on cheap specialist models over frontier AI AI
- OneDrive Photos Silently Installs on Windows 11 With Biometric Scan Capability, No Removal Path Privacy
- Microsoft Overwhelmed by AI-Discovered Bugs as Anthropic’s Mythos Uncovers Hundreds of Vulnerabilities Security
- Microsoft Launches MAI-Cyber-1-Flash, Still Relies on OpenAI for Hard Tasks AI
- Microsoft Mandates Passkeys in Entra by February 2027, Retires SMS and Voice MFA Security
- Microsoft and Mistral Strike Multi-Billion-Dollar Deal to Build AI Infrastructure Across Europe AI
- BitLocker Zero-Day CVE-2026-50661 Allows Physical Bypass of Encryption Security
- Microsoft’s July Patch Tuesday Sets Record with 570 Fixes, Two Zero-Days Under Attack Security
- GigaWiper Modular Backdoor Combines Disk Wiper, Fake Ransomware, and Spyware Security
- Microsoft Urges Shorter Windows Update Deployment Timelines Due to AI Threats Security
- Microsoft’s Global Device ID Used to Track Hacker, Raising Privacy Concerns Privacy
- Hacker Arrest Reveals Microsoft Windows Global Device ID Tracks Users Security
- Microsoft Accelerates Quantum Security Timeline to 2029 Security
- Microsoft Merges Consumer and Enterprise Copilot into Single Super App AI
- Microsoft launches $2.5 billion ‘Frontier Company’ to embed 6,000 AI engineers inside enterprise clients AI
- Microsoft Removes 119 Edge Extensions Hiding StegoMalware Security
- Quantum Computing Race Could Break Encryption by 2028, US Executive Orders Signal Privacy
- Operation Endgame Disrupts Over 200 C2 Servers and Recovers 27 Million Stolen Credentials Security
- Microsoft shareholders sue over Azure slowdown and AI spending AI
- Microsoft may use DeepSeek V4 as optional model in Copilot Cowork, shifts to usage-based billing AI
- Secure Boot Certificates to Expire June 24 — Update Your Systems Now Security
- Microsoft CEO Nadella Warns of AI Concentration Risk and Calls for Proprietary Learning AI
- OpenAI Launches GPT-6 Amid Rising Competition and Financial Pressure AI
- Microsoft Bans Claude Fable 5 Over Anthropic Data Retention Rules Privacy
- Researcher publishes seventh Windows zero-day exploit hours after Microsoft’s record Patch Tuesday Security
- Microsoft Shatters Patch Tuesday Record With Nearly 200 Fixes and Three Zero-Days Security
- Microsoft GitHub Repositories Compromised with Credential-Stealing Malware Security
- Microsoft Repos Used to Deliver Malware via AI Coding Tools Security
FAQ
What is Microsoft?
Microsoft is a US technology company behind Windows, Azure, Microsoft 365, GitHub and the Copilot AI products. In July 2026 it reported more than 30 million paid Microsoft 365 Copilot seats, up from 20 million in April.
How many flaws did Microsoft fix in September Patch Tuesday?
974, a new record. Two were already exploited: CVE-2026-81963 in the Windows Update Stack and CVE-2026-85880 in Advanced Local Procedure Call. More than 110 are rated critical, and by Ars Technica's count Microsoft has fixed 2,760 flaws so far in 2026.
Does Microsoft track Windows users?
Microsoft assigns every Windows installation a persistent Global Device ID (GDID) that it can associate with third-party services and timestamps. The FBI used it to tie an alleged hacker to a hack despite his VPN. A reinstall of Windows resets it, but Microsoft could link the new ID to the old one through a Microsoft account login.
Is Microsoft ending SMS multifactor authentication?
Yes, in Entra: Microsoft will remove SMS and voice MFA on 1 February 2027. From 1 September 2026 users are prompted to register a passkey at their next MFA challenge. Organizations that cannot move away from telephony must configure a supported telecom provider through the Microsoft Security Store by 30 October 2026.
Is Microsoft Copilot safe?
Copilot has been hijacked by prompt injection in research. Researcher Håkon Måløy built a self-spreading worm hidden in Word documents, Microsoft confirmed the behavior on 31 March, two fix attempts failed, and he published after 144 days with no fix in place.
Does Microsoft build its own AI models?
Yes. Microsoft AI builds the MAI models, including MAI-Cyber-1-Flash, which scores 96 percent on the CyberGym benchmark. It still routes hard tasks to OpenAI's GPT-5.4, and its Decision-1 model is based on Qwen3.5-9B.