TRENDS, THREATS & TACTICS FOR CYBER CERTAINTY™
BY DANIEL TOBOK
October 2026
NEW ALL-TIME RECORD LEVELS OF CYBER THREAT AND CONCERN: Global Trends & Threat Levels At An All Time High
A I – O R C H E S T R A T E D P A P E R C U T A T T A C K S : 395 Organizations Compromised
- PaperCut Attacker Uses Hundreds of AI Agents to Compromise 440+ Instances
A suspected Russian-speaking attacker deployed hundreds of autonomous AI agents using OpenAI’s Codex framework with a DeepSeek model to exploit two critical flaws in PaperCut NG/MF print management software — CVE-2026-81578 and CVE-2026-82078 — and compromise at least 440 servers belonging to 395 organizations across 48 countries. The campaign, documented by GreyNoise, achieved remote code execution in under four hours, domain administrator access in six hours, and compromised 11 organizations in 26 seconds once fully launched. Domain administrator access was confirmed in 12 victim organizations, with successful escalations ranging from 5 to 144 minutes. Most victims were schools and universities in the United States, United Kingdom, Canada, and Australia.
My thoughts:
- What GreyNoise documented with the PaperCut campaign is a line being crossed that security professionals have been dreading. Not AI-assisted attack development, which has been happening for years, but a confirmed case where AI agents played a central role across vulnerability research, exploit development, target selection, execution, troubleshooting, and repeated attack attempts. AI-supported workflows can accelerate exploitation across many targets. That changes the calculus on how fast attacks can move.
The number that matters most in this campaign is 26 seconds. Eleven organizations compromised in 26 seconds demonstrates the speed and scale of this AI-enabled campaign. The implication for defenders is direct: human-paced detection and response is structurally too slow against this attack model. The window between initial access and domain compromise was as short as seven minutes at one US high school.
What can we do?
- Patch PaperCut immediately if you have not already. CVE-2026-81578 and CVE-2026-82078 are being actively exploited and PaperCut has released fixes for both. If your organization runs PaperCut NG or MF in any capacity — including in schools, universities, or distributed office environments — treat this as an emergency patch deployment, not a scheduled update. Verify patch deployment to every instance, including any that may have been missed by automated update processes.
Review your detection and response timelines against the reality this campaign establishes. If your mean time to detect is measured in hours and AI-orchestrated attacks can achieve domain compromise in as little as five minutes, that gap is your most urgent operational problem. Run a tabletop exercise focused on how quickly your team can detect and isolate a credential attack moving laterally within minutes of initial access.
Business leaders should understand that this campaign marks the beginning of a permanent shift in the threat landscape. The barrier to conducting attacks at this scale is becoming significantly lower as commercial AI tools and known vulnerabilities are combined. Every organization running unpatched internet-facing software is a potential target. Ask your security team not whether you are patched, but how quickly you would know if an AI-orchestrated attack had already begun moving through your environment.
M I C R O S O F T P A T C H E S A R E C O R D 9 7 4 V U L N E R A B I L I T I E S : Two Exploited Zero-Days Included
- Microsoft Patches Record 974 Flaws, Including Two Exploited Windows Zero-Days
Microsoft released its largest Patch Tuesday on September 8, addressing 974 vulnerabilities across its software portfolio including 723 Windows flaws, 111 in Office, and 62 in SQL Server. Two vulnerabilities were already being exploited before Microsoft released patches: CVE-2026-85880 and CVE-2026-81963, both privilege escalation flaws allowing local attackers to reach SYSTEM-level access. The release also includes 22 critical Office-related vulnerabilities, 12 of which can be triggered through the Preview Pane or Reading Pane without the user opening the file. Microsoft has now patched 2,760 security flaws in 2026 alone, more than doubling its previous annual record of 1,250 set in 2020.
My thoughts:
When I look at 974 vulnerabilities in a single month, my concern is not surprise at Microsoft — it is concern about the organizations receiving them. Nearly one thousand vulnerabilities addressed in thirty days is not a routine maintenance cycle. The drivers are clear: AI-assisted vulnerability research is finding flaws faster than traditional discovery ever could, and the 2,760 CVEs patched so far this year confirm this is not a one-month anomaly.
The two exploited zero-days — CVE-2026-85880 and CVE-2026-81963 — are privilege escalation flaws, which means an attacker who has already gained initial access to a Windows machine can use them to reach full SYSTEM-level control. That is the step that turns a limited foothold into complete ownership of a device. It is also the step that makes lateral movement and ransomware deployment dramatically easier. Both CISA and Microsoft have flagged these as requiring immediate patching.
A user does not have to open a file, run a macro, or click anything unusual. Previewing is enough. That eliminates one of the most commonly repeated pieces of end-user security advice — do not open files from unknown senders. Organizations relying on user behaviour as a control against Office-based malware delivery need to revisit that assumption.
What the volume of this Patch Tuesday tells me strategically is that the vulnerability discovery pipeline has permanently accelerated. AI tools are finding flaws at a rate human researchers cannot match, meaning the time between a vulnerability existing and being exploited is shortening. The organizations that manage this well are not the ones patching everything fastest — they are the ones with the best prioritization process.
What can we do?
Apply the September 8 Windows updates immediately and prioritize CVE-2026-85880 and CVE-2026-81963 above everything else in this release. Both are actively exploited privilege escalation flaws that can be combined with initial access obtained through any other vector to achieve full system compromise. Validate deployment to every Windows endpoint in your environment, including devices that may have been offline or skipped the automated update cycle, and confirm through your endpoint management platform that the patch is applied rather than merely scheduled.
Address the Preview Pane vulnerabilities through configuration rather than user behaviour. Disable the Preview Pane in Outlook and File Explorer as an immediate control while patches are being deployed, particularly for users who regularly receive files from external senders. This configuration change takes minutes and eliminates the zero-click exploitation path for the 12 critical Office flaws. Do not rely on users to avoid previewing files — that is not a realistic control.
For business leaders: the right question after a 974-vulnerability Patch Tuesday is not “are we patched?” It is “how are we deciding what to patch first?” At this volume, patching everything immediately is not operationally realistic. What you need is risk-based prioritization: Prioritize the two actively exploited zero-days first, then assess the remaining vulnerabilities based on exposure, exploitability, business criticality, and affected systems. If your team cannot explain that prioritization clearly, that is the gap to close.
P A S S K E Y – T H E M E D P H I S H I N G : Attackers Hijack Microsoft Cloud Accounts
- Attackers Use Passkey Phishing to Hijack Microsoft Cloud Accounts and Exfiltrate Data
Microsoft disclosed two separate campaigns: one involving CEO impersonation emails aimed at financial fraud, and another using passkey-themed social engineering to compromise Microsoft cloud accounts and exfiltrate data. In the first, over one million financial fraud emails were sent between August 3 and 5 by impersonating CEOs through abused third-party email infrastructure. In the second, attackers used fake passkey enrollment prompts to steal Microsoft cloud credentials and then exfiltrate data from the compromised environments. The passkey phishing technique presents users with what appears to be a legitimate Microsoft passkey setup page. The fake workflow guides victims through credential-capture, adversary-in-the-middle, or device-code authentication methods that give attackers access to the account. Both campaigns have been linked to a cybercrime collective tracked by the security community under multiple aliases.
My Thoughts:
- What Microsoft has documented in the passkey phishing campaign deserves careful attention: attackers are targeting the authentication method the industry has been promoting as the solution to phishing. Passkeys were designed to be phishing-resistant. In the traditional sense, they are. But this campaign demonstrates that the social engineering layer around passkey enrollment is itself a new attack surface that most users are not prepared to recognize.
The phishing flow is precise. An employee receives what appears to be a Microsoft notification to set up a passkey. The page looks authentic. They follow the prompts and hand the attacker a credential granting access to the Microsoft cloud environment, from which data exfiltration proceeds. The attack does not exploit a flaw in passkey technology. It exploits the user’s inability to distinguish a genuine Microsoft passkey enrollment page from a convincing imitation.
The CEO fraud component — over one million emails impersonating CEOs in a two-day window — is a reminder that scale and speed now define financially motivated cybercrime. Abusing legitimate third-party email infrastructure allowed these messages to pass reputation-based filters. The combination of volume, surface-level sender legitimacy, and authority-based social engineering describes an attack model that will become more prevalent as AI lowers the cost of generating convincing fraud content at scale.
What both campaigns share is an exploitation of user trust in familiar interfaces and legitimate-looking communications. Passkey phishing works because most users do not have a mental model for how passkey enrollment can be faked. CEO fraud works because most employees defer to apparent authority. The attack surface in both cases is not technology — it is the human expectation that what looks legitimate is legitimate. That expectation is now being engineered against at industrial scale.
What can we do?
- Implement Conditional Access policies in Microsoft Entra that require device compliance and location verification before passkey enrollment is permitted. Users should not be able to complete a passkey setup flow from an unmanaged device or an unusual location without additional verification. Passkey-themed lures can lead to session theft or unauthorized device-code authentication. Protect authentication-method registration, restrict unnecessary device-code flows, and investigate suspicious sign-ins.
Run a search across your Microsoft 365 and Entra environment for any passkey enrollment events that occurred outside of your standard onboarding workflow or IT-managed enrollment process. Unauthorized registration of a passkey, phone number, authenticator application, or other MFA method can provide attackers with persistent access — once an attacker enrolls their own passkey to a legitimate account, they have a durable authentication credential that survives password resets. Any unexpected enrollment events should be investigated as potential compromise indicators and revoked immediately.
For business leaders: the passkey phishing campaign is a direct challenge to the assumption that moving to modern authentication methods solves the phishing problem. It does not. Phishing evolves to target whatever authentication method your organization uses. The permanent answer is not a specific technology — it is a security culture where employees know to verify unusual authentication requests through a separate channel before completing them, regardless of how legitimate the prompt appears.
