Since last Friday..
The patch window collapsed from weeks to hours — twice in one week, exploitation followed a public disclosure by hours, and in one case it followed the hotfix too. The edge appliance and the remote-management tool were the two multipliers. The AI stopped assisting the attack and started running it. And the phone network, the identity layer, and the vendor you stopped using all turned out to be the target.
Six patterns in this weekly recap: the patch window measured in hours, the edge and the RMM, the AI doing the operation, the phone network and identity, the breach wave with enforcement hitting back, and the policy shelf.
The patch window is now measured in hours
Microsoft’s November Patch Tuesday fixed 1,094 vulnerabilities — 1,235 counting Edge — with 63 rated critical. The second month in a row over a thousand. The one that matters most is the Outlook zero-day, CVE-2026-124194: a crafted file, opened, leaks NTLMv2 hashes without the external-content warning that was supposed to protect you. CISA added it to KEV on November 10, federal deadline November 24. A thousand-plus patches a month is the new normal, and the normal is the problem. The volume means the triage is the job now, not the patching.
The week’s sharpest lesson came from N-central. Researcher Sean Bales published proofs of concept on November 12, and attackers had an XMRig Monero miner running on 21 N-central servers the same day — about five hours later. Nineteen of those 21 servers were already running the latest hotfix, 2026.4 Hotfix 2. The chain still worked. N-able has taken its hosted N-central offline and is telling customers to disable the Legacy API. The second flaw set, Agent2Shell, would let an attacker run commands as SYSTEM on managed Windows endpoints. No fix released for any of it as of November 13.
The same pattern, smaller scale, ran all week. OpenSSH’s SSHole, CVE-2026-123158, lets an attacker reach password-disabled accounts on some Linux systems; it was exploited hours after disclosure and landed in CISA KEV. Google patched its eighth actively exploited Chrome zero-day of 2026 — a use-after-free in the XML component, CVE-2026-125350. A public proof of concept dropped on November 10 for SignDrop, an unpatched Windows privilege escalation abusing CLFS, with no CVE and no patch. And CISA added 11 entries to KEV in the last seven days — BroadWorks, Chrome, Outlook, both Barracuda flaws, Openwave, Aspera, OpenSSH, SignalWire, GPUStack, and SonicWall.
The takeaway is the one the week kept restating: assume exploit. The disclosure-to-exploitation window is hours, the hotfix is not a guarantee, and the triage order matters more than the patch count.
The edge and the RMM are the two multipliers
Fortinet shipped 89 advisories in a single November batch — around 100 vulnerabilities, roughly 17 of them critical. The one with a name is Fortinet2Shell, CVE-2026-121529, CVSS 9.1: a CORS and origin-validation flaw in a FortiClient local service that lets a malicious website run code on the machine, just by being visited. No workaround, no virtual patch.
The Check Point story was already exploited. An unknown actor chained PHP2Shell with a Gaia OS privilege-escalation zero-day, CVE-2026-123039, to get root on Security Gateways — then webshells, a persistent backdoor called TSGA, and in some victims a lateral move to VMware ESXi hosts, where a wiper overwrote files with zeros. Citrix patched 28 NetScaler vulnerabilities — nine critical, plus a tenth in Console, including one that hits all deployments with no workaround. Apache patched a critical mod_http2 remote code execution flaw, CVE-2026-119618. Two Barracuda ESG flaws could be combined for full root access. SonicWall patched an exploited SonicOS flaw that weakens VPN logins, CVE-2026-19868. Cisco BroadWorks carried an actively exploited XXE, CVE-2026-20320. GitLab’s CI/CD regex parser flaw, CVE-2026-97229, made KEV. Red Hat Quay had a critical flaw, CVE-2026-123710, that lets an unauthenticated attacker read and write private container images. Xen Orchestra carried a critical authentication bypass, VSA-2026-062, actively exploited — a crafted cookie logs the attacker in as an admin, with full XCP-ng host takeover as root. No CVE assigned yet.
The second multiplier is the remote-management tool. Huntress tracked a campaign abusing cloud-hosted ScreenConnect to deploy rogue clients and the SCKit toolkit — around six MSPs, 288 endpoints, 48 organizations, in through a third-party extension using a blank-key flaw. N-central is the same story at larger scale: the tool that manages everything is the thing that has to be patched first, and this week it was the thing that got exploited first.
And the retired vendor is the third multiplier. Wiz published the Qubole supply-chain attack: between October 14 and 26, compromised Qubole IAM users assumed customers’ cross-account roles in AWS — roles many customers had left active after they stopped using Qubole. Secrets were pulled from Lambda environment variables, Secrets Manager, and SSM. Entire S3 buckets were downloaded. The IAM role outlives the relationship, and the role is the access.
The AI is doing the operation now
Unit 42 published Cairn — an open-source AI pentesting framework, run by Chinese-speaking attackers with Chinese models, to carry out 25 largely autonomous operations against at least 19 organizations in August and September. Ten ended in unauthorized access or data retrieval. The operators supplied a target and a goal. Cairn planned, executed, and recorded on its own. The targets: semiconductor supply chain, defense policy, civil society. One organization lost more than a million personal records. A live ballot on an industry standard got overwritten.
A jailbroken DeepSeek agent ran a two-week attack on a European telecom, entering through the Cisco BroadWorks flaw above. Symantec detailed an attack on an unnamed US retailer in September: hand-typed commands were blocked for about two hours, then the attacker switched to what looks like AI-generated PowerShell and broke through, harvesting credentials from SolarWinds Orion config files and moving laterally in about ten minutes. The defender blocked the human for two hours. The model got through in ten minutes.
The deception side kept pace. The DOJ announced the FBI disrupted FaceStealer, a North Korean scheme tied to more than $400 million in crypto theft over 18 months, built on fake video meetings with prerecorded webcam clips and AI deepfakes of crypto figures. Arctic Wolf linked the AntiFox AI pentesting platform to attacks deploying webshells and pivoting into AWS production secrets. Microsoft Research showed that editing fewer than ten dormant neurons in an AI agent’s weights can plant a backdoor — no retraining, and the models still pass standard evaluations.
The defense side is finding faster than it is fixing. VulnCheck audited Anthropic’s Project Glasswing — the program where AI finds vulnerabilities in open-source software — over seven months: 37,966 findings, only 217 fixed. Since September: 11,813 new findings, 15 fixes. OpenAI released GPT-6 Cyber on November 5 — a model built to find vulnerabilities and turn them into working exploits, rated “Critical” for cybersecurity, gated behind Daybreak Red with identity checks and IP allowlisting. Fewer than 25% of the reported vulnerabilities have been fixed, by OpenAI’s own dashboard. OpenAI launched Daybreak on November 12 — Red Team agents that pen-test web apps, and Defense Factory, which automates the find-validate-fix loop and drafts merge-ready patches for engineer approval. Discovery is now happening at machine speed. Remediation isn’t. Recorded Future tracked 103 AI-linked incidents in a year, with credential theft through compromised AI tool packages. The AI coding assistant holds your credentials, your keys, your repository access. The package that compromises it inherits all of it.
The phone network and the identity layer
Enea named a new state-level actor, SilentAir, hijacking nodes inside Asian mobile operators to launch SS7 and Diameter signaling attacks over the IPX interconnect. Attack sources sit in at least six Asian countries; Taiwan is the primary target. Two-factor codes were seen being intercepted moments after the diversion began. The phone network is the identity layer now. The signaling protocol is the attack surface.
The same week, a Latvian telecom left a ClickHouse database exposed with more than 125 million SMS messages inside — including messages for Swedbank, SEB, and airBaltic customers. No intrusion, no agent, no exploit. A misconfiguration, open for a month. And Turla’s Cereal toolkit turned Exchange servers into spy hubs, collecting email from government systems with no external command and control — the collection stays inside. Proofpoint’s telemetry has phishing detections up 4.5x year over year across Europe, against 1.5x globally — and Europeans clicked 8.3 per 100 delivered threat messages, against 2.0 globally.
The identity layer showed its seams in the breach reports too. Apify told customers to reset two-factor authentication, because 2FA codes may have been exposed after a compromised credential opened its internal systems. Microsoft’s own outage hit the security tooling: Defender XDR portals and Defender for Office 365 Safe Links were degraded during the Microsoft 365 outage, which may not have been able to verify URLs — the phishing-risk angle during recovery. Microsoft’s Entra Hard Deletion Protection went into preview: once enabled, a deleted identity can’t be permanently erased for seven days. Seven days to recover the identity you deleted by mistake — or the one someone deleted on purpose.
The breach wave, and enforcement hitting back
Maersk confirmed unauthorized access and data extraction from a legacy LF Logistics system, with WhiteArmor claiming around 100GB and demanding $5M. Baker Tilly is notifying clients that personal information was compromised in a breach at a third-party platform. TSMC confirmed about 46,000 employee records leaked after a hack of an external cloud service provider. Nitori, Japan’s largest furniture retailer, disclosed that an attacker may have taken up to 37.62 million customer records, with signs of access since mid-October. Sega’s arcade network had a security incident. LEGO’s BrickLink marketplace went dark for a week on unusual activity, then came back with usernames and IP addresses accessed. Amplifon Americas disclosed a breach affecting 1,332,826 people — discovered eight months after the files were taken.
The credential pipeline underneath the breaches ran all week. Aikido found wfx-loot: 4,837 hijacked GitHub accounts planting probe pull requests across 61,764 repositories, with the stealer running in more than 1,806 of them. Fake GitHub repositories posing as security utilities infected more than 2,000 Windows machines with AsyncRAT and a stealer. Truffle Security scanned 224 million public GitHub repositories and found 543,699 unique credentials that still authenticated — the median one sitting in a public default branch for 784 days. The oldest was committed in 2009 and still works. And Arkham tracked 1,052.34 bitcoin — about $78.61 million — leaving wallets labeled as belonging to the Bitget hacker, most of it through 66 CoinJoin transactions. The theft was September. The laundering is now.
Enforcement answered in kind. Europol and Eurojust announced the latest Operation Endgame action: 235 servers and 257 domains seized, 6.4 million victim credential datasets identified — Gootloader, CastleRAT, and StealC. Gootloader alone had infected at least 25,000 computers since 2018. No arrests announced. The infrastructure is gone. The credentials are still out there. And Operation TorNado: 354 arrests and more than $234 million in cash and crypto seized, across 14 countries — the largest JCODE operation to date, targeting vendors and buyers on five already-seized dark-web markets.
The policy shelf
Japan’s prime minister announced a ¥10 billion emergency package — about $63 million — on November 6, with an AI “top gun” expert team by the end of November and a public-private cyber rescue team as first responder. The context: more than 600 incidents publicly reported in the first nine months of 2026, against 594 in all of 2025.
CISA’s 2030 strategy — Defend, Harden, Prevail — puts threat disruption and risk-prioritized hardening front and center. Congress is back with a December 11 deadline: the Cybersecurity Information Sharing Act of 2015 expires again that day, along with the Federal Cybersecurity Enhancement Act and the Technology Modernization Fund. The protections that make companies share threat data expire in a month. The UK’s NCSC launched its Cyber Adversary Simulation scheme on November 11, vetting providers of adversary simulation for critical organizations. The UK’s Cyber Security and Resilience Bill cleared the Lords on November 2 and has to pass by April 2027 or it lapses. ENISA’s 2026 State of Cybersecurity report puts the EU Cybersecurity Index at 71.53 out of 100 — supply-chain management scores 89.91, CSIRT maturity sits at 54.76 and is uneven across member states.
And the people moved. Jensen Huang stepped down as NVIDIA CEO on November 12, days after reports of an affair with an executive at Singapore-based Megaspeed, whom US authorities suspect of spying for Beijing — and Congress is moving on chip export controls in response. The Senate set a hearing for Thursday, November 19, to consider Michael LoSapio’s nomination as CISA director. CISA has had no Senate-confirmed director since January 2025.
The takeaway
The week’s through-line is the compression of every window that used to buy time. Disclosure to exploitation: hours. Finding to fixing: a gap of tens of thousands. The hotfix that was supposed to close the window didn’t. The AI that was supposed to help the defender is now running the operation for the attacker, and the defense’s AI is finding flaws faster than anyone can fix them. The phone network, the port, the cloud provider holding the employee data, the vendor you stopped using — the thing you assumed was infrastructure turned out to be the target. The patching is still on us. The finding, increasingly, is not. And the one window that hasn’t compressed is the policy one: the laws and the leadership that are supposed to catch up are still working on December deadlines and January confirmations.
Leave a Reply