#podcast

Weekly Security Roundup — week ending 15 November 2026

AI attack tools go operational (Artex, ArcticRAT); Fireworks AI breach hits Cursor and Notion; Windows SignDrop zero-day; Check Point ESXi wipers; FAA ground stop (single router); Trezor hardware implants; Gootloader takedown; NCSC AI threat assessment to 2028.

Security Stories

AI as an Offensive Weapon — Artex, ArcticRAT, Autonomous AI

The FBI issued a flash alert on Artex, an open-source AI-powered penetration testing tool weaponised by criminal actors, reducing intrusion chains to under three hours from initial targeting to privileged access. Cisco Talos and Unit 42 published coordinated research the same day. Dark Reading reported on ArcticRAT, which uses AI to profile victims' financial tolerance before ransomware deployment. Unit 42 documented autonomous AI targeting semiconductor, defence, and civil society sectors — reconnaissance and targeting with no human intervention between steps.

Fireworks AI Breach Cascades — Cursor, Notion, GitHub

Fireworks AI, an AI infrastructure provider, was breached — and the cascade reached two downstream services. Cursor confirmed attackers used stolen Fireworks credentials to push a malicious Claude Code extension into its marketplace. Notion confirmed customer AI chat transcripts were accessed. Attackers also compromised approximately 4,800 GitHub accounts to steal cloud credentials. Microsoft research detailed post-training supply chain attacks that corrupt AI coding agents after training. The SCKit worm campaign spread through MSPs via a compromised ScreenConnect plugin.

Windows SignDrop Zero-Day + Project Zero Hard Deadline + CVE 100K

Nightmare Eclipse dropped SignDrop, a fully working Windows zero-day exploiting the Common Log File System driver and Windows Defender to escalate to SYSTEM privileges. No patch has been issued. Google Project Zero announced a hard 90-day disclosure deadline with no vendor extensions permitted. The annual CVE count passed 100,000 — more than double 2025's total.

Check Point Zero-Day with ESXi Wipers + N-central + Apache RCE

Check Point patched an actively exploited zero-day in Security Gateway, with PHP2Shell attacks deploying VMware ESXi wipers — destructive malware targeting virtualised infrastructure. Emergency hotfixes are available. N-able took N-central offline after finding 33 vulnerabilities, one exploited hours after PoC publication. Apache patched a critical HTTP/2 RCE found jointly by Anthropic and Google.

FAA Nationwide Ground Stop — Single Router, Not a Cyberattack

The FAA issued a nationwide ground stop — only the third in US aviation history — causing 5,000 Friday cancellations, 10,000 delays, and 4,000–5,000 more Saturday cancellations. Officials confirmed no cyberattack — the FBI is investigating a software update that caused a router failure in a southern Arkansas switching facility. The incident prompted scrutiny of SMART, the FAA's recently expanded AI traffic management system, which was not implicated. Separately, Microsoft suffered a global outage affecting Outlook, Exchange Online, and Teams.

Trezor Hardware Supply Chain Attack

Trezor confirmed three Safe 5 cryptocurrency wallets sold by Southeast Asian reseller CryptoBilis contained physical hardware implants designed to intercept recovery phrases and transmit them over cellular networks. The implants were inserted before devices reached buyers. Firmware update released but affected hardware units remain compromised.

Law Enforcement Wins — Gootloader, Binance, FaceStealer

Europol dismantled infrastructure behind Gootloader, CastleRAT, and StealC — seizing 235 servers, 257 domains, and 6.4 million victim credentials. DOJ confirmed Gootloader funded 44 ransomware attacks costing $29M since 2018. The FBI disrupted North Korea's FaceStealer campaign ($400M+ in crypto theft) and seized $1.2B in Binance-stolen crypto. Europol arrested 354 suspects in a dark web crackdown.

Megaspeed — Alleged China Nvidia Export Control Bypass

A CNN investigation raised serious questions about Megaspeed International, which allegedly went from dormant Chinese gaming subsidiary to one of Asia's biggest Nvidia GPU buyers in three years. The report suggests a possible Chinese state-sponsored network for bypassing US export controls — intersecting with the broader Jensen Huang/Nvidia governance crisis this week.


Key Themes

AI has crossed the threshold from theoretical to operational offensive capability. Artex automating intrusion chains, ArcticRAT profiling ransom tolerance, Unit 42's autonomous AI, and the NCSC's new threat assessment (see below) all point to the same inflection point. The defenders are still catching up to AI as a tool; the attackers are already deploying it.

The supply chain attack surface now extends to AI and development infrastructure. Fireworks AI → Cursor/Notion, the GitHub credential campaign, SCKit through ScreenConnect, and Microsoft's research on post-training model corruption all target the same thing: the tools that developers and organisations trust. Compromising one upstream provider creates leverage across all its downstream users.

Infrastructure resilience is a vulnerability disclosure problem in disguise. The FAA and Microsoft outages were not cyberattacks, but they demonstrated the same fragility that adversaries exploit. One router. One update. Meanwhile, unpatched Windows zero-days, ESXi wipers, and a CVE count at 100,000 mean the defensive window between disclosure and exploitation is narrowing faster than patch cycles can accommodate.


NCSC — AI and Cyber Intrusion Assessment (3 November 2026)

The National Cyber Security Centre published a major strategic assessment this week: "Artificial intelligence and cyber intrusion", examining current AI-enabled threat trends and projections to 2028. It is written for security professionals and large organisations.

The key judgements use NCSC's probability framework, and the language is unambiguous. By 2028, it is highly likely that Russia, China, and other nation states will routinely use AI-cyber tools to achieve geopolitical goals, increasing the scale and speed of attacks against the UK. It is almost certain that many systems will remain vulnerable to AI-cyber tools that identify and exploit weaknesses — and that attackers will display less caution about burning access, because with AI, replacing access is cheap. AI-powered attacks against UK CNI legacy technology are assessed as likely. And critically: it is highly likely that between now and 2028, AI-powered malware will have been used in a ransomware attack on UK critical national infrastructure.

The report presents three scenarios — a baseline of increased but managed threat, a pessimistic case where AI proliferation far outstrips UK defensive capability (including the emergence of autonomous cyber actors outside operator control), and an optimistic case where development slows enough for robust countermeasures to develop. The five key drivers identified are: geopolitical tensions, UK CNI cyber defence capability, the pace of AI development, the availability of AI-cyber tools in the marketplace, and the state of a cyber-literate workforce. The report does not assign a probability to the three scenarios — but nothing in this week's news points toward the optimistic one.


HaveIBeenPwned — Breach Activity This Week

BreachDate AddedAccountsData Exposed
YouHack11 Nov 20267,160,849Email addresses, IP addresses, Passwords, Usernames
SberSpasibo (Sberbank loyalty)11 Nov 20263,288,711Bank card numbers, Dates of birth, Email addresses, Loyalty program details, Phone numbers
CastleRAT13 Nov 2026179,745Email addresses, Passwords