> ## Content Index
> Fetch the complete content index at: https://www.cyberverso.net/llms.txt
> Use this file to discover other available public pages before exploring further.

# Cyber / Brief, 21 Nov 2026: Chinese hackers' AI agents broke into two universities, Amazon says
- URL: https://www.cyberverso.net/brief/cyber-brief-21-nov-2026/
- Published: 2026-11-21T07:23:19.000Z
- Updated: 2026-11-21T07:23:19.000Z
- Description: Amazon says AI agents run by an operator writing in Simplified Chinese found an unknown weakness in a medical-imaging tool and used it to break into two East Asian universities without a human choosing each step, as Bell Canada blamed an "agentic" attacker for stealing enterprise customer…
- Author: Paolo De Rosa
- Tags: #bulletin, #signal, #en

Amazon says AI agents run by an operator writing in Simplified Chinese found an unknown weakness in a medical-imaging tool and used it to break into two East Asian universities without a human choosing each step, as Bell Canada blamed an "agentic" attacker for stealing enterprise customer data and SOCRadar showed a low-skilled operator turning a jailbroken agent on banks and municipalities in Belgium, France and Switzerland. In Washington, two House committees warned US companies off Chinese open-weight models and proposed a federal American alternative, while Trump threatened a heavy tax on US firms that build data centres abroad, a direct question for Europe's plans to host AI compute. Berlin says European investors could put €25 billion into buying TikTok's EU business, and the EU's data protection supervisor warned that the Kids Act would make age checks routine across everyday online services.

## Top Stories

- [US companies’ use of Chinese open weight AI draws rebuke from House chairmen](https://therecord.media/chinese-open-weight-AI-use-by-US-companies-house-investigation) — *The Record from Recorded Future News* · AI & Power
- [Amazon says Chinese hackers used Cairn AI tool to discover apparent zero-day during attack on university](https://therecord.media/amazon-research-cairn-ai-university-attacks-possible-zero-day) — *The Record from Recorded Future News* · Threat Intelligence (CTI)
- [Signal blocks Apple, Google AI screenshotting in response to ‘existential threat’ to privacy app](https://cyberscoop.com/signal-blocks-apple-google-ai-screenshots/) — *CyberScoop* · AI & Power
- [Researcher plants 2m canaries into the AI supply chain. A privacy disaster surfaced.](https://arstechnica.com/security/2026/11/a-leaky-ai-supply-chain-is-exposing-secret-user-data-at-an-alarming-rate/) — *Ars Technica* · AI & Power
- [Germany says investors ready to spend €25B to buy TikTok Europe](https://www.politico.eu/article/germany-says-investors-ready-to-spend-e25b-to-buy-tiktok-europe/?utm%5Fsource=RSS%5FFeed&utm%5Fmedium=RSS&utm%5Fcampaign=RSS%5FSyndication) — *Technology – POLITICO* · EU & Technology

---

## AI & Power

[US companies’ use of Chinese open weight AI draws rebuke from House chairmen](https://therecord.media/chinese-open-weight-AI-use-by-US-companies-house-investigation) — *The Record from Recorded Future News*  
Why it matters: Congress treats Chinese open-weight models inside US companies as the next Huawei problem.  
A 50-page joint report by the House Homeland Security Committee and the Select Committee on the CCP, led by Reps. Andrew Garbarino and John Moolenaar, names Airbnb, DoorDash and Cursor owner Anysphere as US firms that built Chinese open-weight models into customer service, coding and inventory tools. It argues that, as with Huawei and ZTE, removing such models later would take years and cost billions, and cites federal findings that Chinese developers used proxy networks and fraudulent accounts to harvest outputs from US frontier models.

[Signal blocks Apple, Google AI screenshotting in response to ‘existential threat’ to privacy app](https://cyberscoop.com/signal-blocks-apple-google-ai-screenshots/) — *CyberScoop*  
Why it matters: Signal blocks the AI agents built into iOS and Android from reading its users' messages.  
Signal will by default stop Apple's and Google's operating-system assistants (Siri's Onscreen Awareness and Gemini Intelligence) from capturing its screens, extending the block it applied to Microsoft Recall last year. Foundation president Meredith Whittaker called OS-level agents an "existential threat" to the app and asked Apple, Google and Microsoft for a clear way for developers to opt out of AI data access. The move follows reports that Meta's Muse agent could read Signal messages.

[Researcher plants 2m canaries into the AI supply chain. A privacy disaster surfaced.](https://arstechnica.com/security/2026/11/a-leaky-ai-supply-chain-is-exposing-secret-user-data-at-an-alarming-rate/) — *Ars Technica*  
Why it matters: Two million planted secrets show prompts leaking across AI routers and resellers.  
Researcher HD Moore seeded unused AWS keys, SSH passwords and unique URLs, one per prompt, into requests sent through AI routers and inference providers such as OpenRouter and Vercel, then watched which ones were used. He found at least three providers compromised by attackers siphoning prompts, resellers that logged prompts and passed them on to people who used the credentials, and providers ignoring zero-data-retention flags, Ars Technica reports.

[Trump, Xi agreed that military AI decisions should involve humans, Bessent says](https://www.defensenews.com/global/asia-pacific/2026/11/20/trump-xi-agreed-that-military-ai-decisions-should-involve-humans-bessent-says/) — *Defense News*  
Why it matters: Washington and Beijing agree that humans must stay in military AI decisions, per Bessent.  
Treasury Secretary Scott Bessent told CNBC that Trump and Xi agreed in Shenzhen that humans should always be involved in military decisions using AI and that it should not act autonomously in major military operations, and called it one of the trip's main results. He said the H200 remains the most advanced Nvidia chip China can buy under licence and that Chinese orders for it have been scarce. No text of the agreement has been published.

[OpenAI tightens grip on model resellers](https://www.semafor.com/article/11/20/2026/openai-tightens-grip-on-model-resellers) — *Semafor*  
Why it matters: OpenAI puts the routers that resell its models under its own approval.  
Gateways and routers that give third parties access to OpenAI's models, OpenRouter the best known, now need OpenAI's review and approval, Semafor reports. US agencies and labs including Anthropic have accused Chinese labs of distilling American models through intermediaries that hide the end user; the rule extends OpenAI's oversight from gray-market "transfer stations" to every service between it and its users.

[Combatting adversarial distillation](https://openai.com/index/combatting-adversarial-distillation) — *OpenAI News*  
Why it matters: OpenAI details how it detects and blocks attempts to copy its models through their outputs.  
OpenAI says it now treats adversarial distillation, training a rival model on its outputs without authorisation, with safeguards close to those it uses for biology and cyber, combining red-teaming, classifiers, API controls and investigations. It focuses on protecting the models' hidden reasoning traces: in its own adaptive test, the success rate of attacks recovering that reasoning fell from 54% against GPT-6 Astra to 0.2% against GPT-6.2 Astra. The figures are the company's own.

[The most misunderstood graph in AI just got a major update](https://www.technologyreview.com/2026/11/20/1149399/the-most-misunderstood-graph-in-ai-just-got-a-major-update/) — *MIT Technology Review*  
Why it matters: METR's updated time-horizon chart: coding keeps doubling, judgement tasks lag.  
The AI evaluation nonprofit METR updated its plot of the length of tasks frontier models can complete, which previously showed that length doubling every four months. The new suite adds longer and more open-ended tasks, and MIT Technology Review reports that exponential gains continue on verifiable work such as coding while models do less well on tasks that require judgement, where success is harder to check.

[Dutch data watchdog highlights safety issues with AI glasses](https://www.theregister.com/security/2026/11/20/dutch-data-watchdog-highlights-safety-issues-with-ai-glasses/5311345) — *www.theregister.com - Articles*  
Why it matters: The Dutch data protection authority advises consumers not to buy AI smart glasses.  
The Autoriteit Persoonsgegevens tested six models priced from under €100 to over €500 and found glasses that misread medication leaflets and toy safety warnings, gave dating advice to a wearer who expressed interest in a minor, and in a CPR simulation claimed emergency services had been called when they had not. It did not name the models. The regulator had already warned in September about covert filming and GDPR risks.

[Anthropic mathematician claims to have solved prestigious Millennium Prize problem with Claude](https://www.theverge.com/ai-artificial-intelligence/1010051/anthropic-riemann-hypothesis-millennium-prize-problem) — *The Verge*  
Why it matters: An Anthropic mathematician says he proved the Riemann hypothesis with Claude; it is unverified.  
Levent Alpöge says he has, working with Claude, found and formally verified a proof of the Riemann hypothesis, one of the Millennium Prize problems, and that the method extends to the generalised version. The proof has not been checked independently and Anthropic has not commented; Alpöge writes that he held publication until after the company's IPO on Thursday, The Verge reports.

---

## EU & Technology

[Germany says investors ready to spend €25B to buy TikTok Europe](https://www.politico.eu/article/germany-says-investors-ready-to-spend-e25b-to-buy-tiktok-europe/?utm%5Fsource=RSS%5FFeed&utm%5Fmedium=RSS&utm%5Fcampaign=RSS%5FSyndication) — *Technology – POLITICO*  
Why it matters: Berlin says European investors could put €25 billion into a forced sale of TikTok Europe.  
German culture minister Wolfram Weimer said investors in five European countries could invest about €25 billion and that TikTok Europe should sell 75% to 80% of its shares, modelled on the US divestment led by Oracle. Germany and Austria floated the idea this week over TikTok's influence on European democracy; TikTok called it "nonsense", saying it already falls under the DSA and the GDPR. POLITICO's owner Axel Springer has said it would be open to taking part.

[Kids Act: EU-Datenschutzbehörde warnt vor Alterskontrollen](https://netzpolitik.org/2026/kids-act-eu-datenschutzbehoerde-warnt-vor-alterskontrollen/) — *netzpolitik.org*  
Why it matters: The EDPS warns that the Kids Act would spread age checks across everyday online services.  
In a 27-page opinion of 12 November, the European Data Protection Supervisor Wojciech Wiewiórowski supports EU-wide child protection online but says the Commission's planned age gates reach far beyond adult content, to social media, video platforms, AI chatbots, games and apps deemed risky, and would in effect make age verification routine for services people need in daily life. He recommends fewer age barriers and safer default features for all users, as Parliament and Council prepare their positions, netzpolitik reports.

[‘Staff deserve answers’ on Commission cyberattack, union says](https://www.politico.eu/article/staff-deserve-answers-eu-commission-cyberattack-union-says/?utm%5Fsource=RSS%5FFeed&utm%5Fmedium=RSS&utm%5Fcampaign=RSS%5FSyndication) — *Cybersecurity and Data Protection – POLITICO*  
Why it matters: Commission staff learned of the AI-assisted cyberattack from the press, their union says.  
Cristiano Sebastiani, who chairs the Commission branch of the Renouveau & Démocratie union, wrote that the executive has yet to send staff a single update on the attack that took HR systems offline, that some managers called it a routine technical issue, and that sensitive personal data may be at risk. The union asks for a full explanation of what data was affected and an independent review; the Commission has informed the European Data Protection Supervisor.

[Mobilfunk aus dem All: Kampf um die digitale Zukunft](https://netzpolitik.org/2026/mobilfunk-aus-dem-all-kampf-um-die-digitale-zukunft/) — *netzpolitik.org*  
Why it matters: Direct-to-phone satellite service is the next telecom market, and the EU is writing its rules now.  
netzpolitik describes how low-orbit constellations from SpaceX, Amazon and others will soon connect ordinary phones for data and voice, a segment estimated at about $100 billion a year by 2034, and argues that whoever controls it will set the terms of mobile communication. Member states are reworking the Commission's satellite services law; Rapporteur reports that France and Spain want provisions sought by Europe's largest mobile operators, who fear competition from SpaceX.

[EU Nears New Trade Power to Block Access to Its Single Market](https://www.bloomberg.com/news/articles/2026-11-20/eu-nears-new-trade-power-to-block-access-to-its-single-market) — *Bloomberg Politics*  
Why it matters: The EU prepares "ReadyEU", a power to shut some goods and services out of the single market.  
The bloc is finalising a tool, provisionally called ReadyEU, that would let it rapidly block some goods and services from its market of 450 million consumers in response to trade attacks from countries such as China, Bloomberg reports, citing people familiar with the plans; it would cover three main cases. Excerpt-based.

[€100B in company value, one shared ambition: European tech leaders demand political reform at Slush](https://tech.eu/2026/11/19/100b-in-company-value-one-shared-ambition-european-tech-leaders-demand-political-reform-at-slush/) — *Tech.eu*  
Why it matters: Thirty-five European founders at Slush ask for capital, market and policy reforms to scale at home.  
Founders and executives representing over €100 billion in company value and 115,000 jobs signed a declaration calling for three political shifts, arguing that Europe lacks political will, not talent or capital: households hold more than €12 trillion in savings while foreign investors capture the growth stage of European deep tech. They urge faster implementation of the Savings and Investments Union, Tech.eu reports.

---

## US & Technology

[Trump threatens ‘tremendous tax’ for moving AI data centers outside the US](https://www.theverge.com/policy/1038892/trump-foreign-data-centers-tax) — *The Verge*  
Why it matters: Trump threatens a "tremendous tax" on US companies that build data centres abroad.  
At a press conference after the Shenzhen summit, Trump said the US will not allow its companies to build data centres outside the country and that those that do will face a heavy tax, adding that he had not yet told his own officials. He gave no details. The threat comes as US states and cities restrict new data centres, which pushes operators to look abroad, and as a Democratic Congress takes office in January, The Verge reports.

[House Lawmakers Push US AI Models to Counter China](https://www.healthcareinfosecurity.com/house-lawmakers-push-us-ai-models-to-counter-china-a-33482) — *HealthcareInfoSecurity.com RSS Syndication*  
Why it matters: The China committee wants a federal open-weight AI programme and a ban on Chinese models in agencies.  
The House Select Committee on China recommends a national American open-weight AI initiative pairing public compute, shared evaluation and lawful access to US frontier models with anchor demand from federal procurement, a ban on Chinese models in federal agencies, consistent testing of those models and sanctions on companies that illicitly distil US models. Before the report, CNBC reported that Cursor, now owned by SpaceX, will stop using Chinese models.

[Phew! Some Regulatory Relief for AI Hacks](https://www.lawfaremedia.org/article/phew!-some-regulatory-relief-for-ai-hacks) — *Articles*  
Why it matters: The White House AI task force asks regulators to go easy on victims of AI-powered attacks.  
In a 13 November statement the task force said AI has changed the risk environment, that prior standards for regulatory action after cyber incidents may be outdated, and that strict enforcement could chill voluntary reporting; the leniency would last at least 12 months. Lawfare's authors back it in the short term but want policy that also gives organisations room to test AI-based defences such as AI red-teaming.

[Outages trigger congressional scrutiny of AI-enabled attacks](https://www.semafor.com/article/11/20/2026/outages-trigger-congressional-scrutiny-of-ai-enabled-attacks) — *Semafor*  
Why it matters: Congress moves on AI-enabled cyber threats as the outages pile up and CISA stays leaderless.  
The administration attributes the recent outages to firms rushing to patch software that AI has exposed, and experts largely agree, Semafor reports. Sens. Maria Cantwell and Mike Rounds asked DNI Jay Clayton's AI task force to launch a national effort against AI-enabled cyber threats, which sources expect to become a legislative priority in January; Sens. Thom Tillis and Ron Wyden will hold CISA nominee Michael LoSapio after his hearing.

[The D.C. Circuit Curbed Pro-executive Panels Before. Is Anthropic’s Case Next?](https://www.lawfaremedia.org/article/the-d.c.-circuit-curbed-pro-executive-panels-before.-is-anthropic-s-case-next) — *Articles*  
Why it matters: Anthropic's challenge to its Pentagon blacklisting may get a full D.C. Circuit rehearing.  
A 2-1 panel upheld the Defense Secretary's immediate exclusion of Anthropic under the Federal Acquisition Supply Chain Security Act; Anthropic asked for rehearing on 13 October and the government opposed it on 18 November. Lawfare argues that the en banc court has repeatedly reversed 2-1 panels favouring executive power since 2018, so a rehearing and a later Supreme Court fight are plausible.

[EFF to Congress: Mass Surveillance Is Too Important to Shove into Must-Pass Bill](https://www.eff.org/deeplinks/2026/11/eff-congress-mass-surveillance-too-important-shove-must-pass-bill) — *Deeplinks*  
Why it matters: Civil-liberties groups warn against renewing FISA Section 702 inside a must-pass bill.  
Section 702 lapsed in June without agreement on its terms and is due to switch off in the coming months. EFF, the ACLU, the Brennan Center and more than 20 other groups wrote to Congress against attaching its renewal to the NDAA or an end-of-year funding bill without a full debate. The programme compels telecom firms to hand over communications of foreigners abroad, which also sweeps in Europeans' exchanges with Americans.

---

## China & Technology

[Tencent releases Hy4 model weights with million-token context](https://technode.com/2026/11/20/tencent-releases-hy4-model-weights-with-million-token-context/) — *TechNode*  
Why it matters: Tencent opens the weights of its 770-billion-parameter Hy4 flagship.  
Tencent released downloadable weights for Hy4, a mixture-of-experts model with 770 billion parameters (49 billion active per token), a context window of about a million tokens and text, image and video understanding, with deployment guides and a safety evaluation card. TechNode reports an Apache 2.0 licence, while Pandaily describes the licence as conditional; the terms are worth checking before use.

[Alibaba uses its own chips to train next-generation Qwen model](https://technode.com/2026/11/20/alibaba-uses-its-own-chips-to-train-next-generation-qwen-model/) — *TechNode*  
Why it matters: Alibaba trains its next Qwen model on its own T-Head chips.  
CEO Eddie Wu told the earnings call that Alibaba is training the next Qwen model on chips from its T-Head unit and that its Bailian (Model Studio) platform already serves inference on Chinese-made chips, extending domestic hardware to both stages of AI computing, TechNode reports, citing CLS. Wu argued the full stack gives Alibaba an edge in China's domestic compute supply chain.

[China Ratchets Up Its Weaponization of Trade](https://thediplomat.com/2026/11/china-ratchets-up-its-weaponization-of-trade/) — *The Diplomat*  
Why it matters: China's export-control system, built on the US model, is now a tool of retaliation.  
The Diplomat traces how Beijing assembled its Unreliable Entities List (2019), Export Control Law (2020), blocking rules (2021), wider dual-use controls and a unified licensing system demanding transaction-level approvals, technical disclosures and end-user checks, mostly in response to US measures, and argues that it now uses them more freely than Washington.

[Baidu fined US$75,000 as China wraps up Joyy antitrust case](https://www.scmp.com/tech/big-tech/article/3375283/baidu-fined-us75000-china-wraps-joyy-antitrust-case) — *Tech - South China Morning Post*  
Why it matters: SAMR closes the Baidu-Joyy case with the old maximum fine.  
China's market regulator fined Baidu's Hong Kong unit 500,000 yuan for closing part of its purchase of YY Live before antitrust clearance, the cap under the pre-2022 Anti-Monopoly Law, while finding the deal did not restrict competition. The case, agreed in 2020 for $3.6 billion and closed in 2025 at about $2.1 billion, was reshaped by Beijing's scrutiny of tech deals, SCMP reports.

---

## Threat Intelligence (CTI)

**\[P1\]** [Amazon says Chinese hackers used Cairn AI tool to discover apparent zero-day during attack on university](https://therecord.media/amazon-research-cairn-ai-university-attacks-possible-zero-day) — *The Record from Recorded Future News*  
Why it matters: Amazon: a China-based operator's AI agents found and used a zero-day to break into two universities.  
Amazon Threat Intelligence found an exposed instance of the open-source Cairn penetration framework, with prompts in Simplified Chinese, set to obtain shell access at East Asian universities; after failed runs (148 actions over 16 hours, a Log4Shell attempt, 30,000 credential guesses) the agents found an unpatched path traversal flaw in an open-source medical-imaging AI application and used it on two universities. The maintainers have fixed it.  
severity high · exploited in the wild · EU: NIS2, AI Act · actor China-based operator (50%), escalation

**\[P2\]** [BabidiGhost: An AI-Augmented Cybercrime Operation](https://socradar.io/blog/babidighost-ai-augmented-cybercrime/) — *SOCRadar*  
Why it matters: A low-skill French-speaking operator used a jailbroken AI agent against banks and towns in Belgium, France and Switzerland.  
SOCRadar reconstructed from an exposed staging server how "BabidiGhost" spent nine days in November directing a jailbroken locally run agent in misspelled French against more than 50 targets, mostly in Belgium, France and Switzerland; of 14 studied, two were breached and three partly, often through open storage buckets and exposed .env files. One compromised server held data for over a dozen Belgian municipalities.  
severity high · exploited in the wild · EU: NIS2, GDPR, DORA · actor BabidiGhost (60%)

**\[P2\]** [AI Joins the Industrial Attack Workflow](https://www.recordedfuture.com/blog/ai-industrial-attack-workflow) — *Recorded Future*  
Why it matters: Recorded Future: the IRGC-linked group behind the French fridge attacks widens its targets and uses AI.  
Recorded Future tracks GreenJuliett (CyberAv3ngers, Bauxite), likely part of the IRGC unit France blamed on 9 November for altering refrigeration controllers at over 700 shops in October; between August and October its infrastructure also probed Fortinet FortiGate firewalls protecting OT networks in six Gulf states and Israel and is adding IoT and CCTV devices, and AI is now part of its workflow. It has not tied the infrastructure to the French attack itself.  
severity high · exploited in the wild · EU: NIS2, CER · actor Shahid Kaveh (IRGC-CEC) (70%)

**\[P3\]** [Suspected North Korea-Aligned Group Hides C2 on Solana Blockchain](https://www.infosecurity-magazine.com/news/overlord-rat-solana-blockchain-c2/) — *Infosecurity Magazine*  
Why it matters: A RAT hides its command servers in Solana transactions and survives a Windows reset.  
Cisco Talos says UAT-12643, assessed with low confidence as North Korea-aligned, stores encrypted C2 addresses in memos of Solana transactions read by its Overlord RAT, which crashes Windows if killed and plants itself in the Recovery Environment to survive a full reset; it also deploys an infostealer and a crypto clipboard hijacker.  
severity medium · exploited in the wild · EU: NIS2 · actor UAT-12643 (possibly DPRK-aligned) (30%)

**\[P3\]** [Ransomware: What we know about the arrests of Chaos affiliates](https://databreaches.net/2026/11/20/ransomware-what-we-know-about-the-arrests-of-chaos-affiliates/) — *DataBreaches.Net*  
Why it matters: Ukraine details four arrests from a 60-strong Chaos ransomware network with Conti roots.  
Ukrainian authorities say the four detainees belonged to a 27-person cell of a 60-member network including former Conti members; under the BlackSuit name they allegedly hit tyre maker Maxxis in 2024 before moving to Chaos. They are held until 1 January.  
severity medium · EU: NIS2 · actor Chaos (ex-BlackSuit/Royal, Conti lineage) (70%)

---

## Digital Sovereignty & Identity

[Poland approves draft EUDI Wallet law, building on mObywatel](https://www.biometricupdate.com/202611/poland-approves-draft-eudi-wallet-law-building-on-mobywatel) — *Biometric Update*  
Why it matters: Poland adopts its draft EUDI Wallet law, built on the 13.5-million-user mObywatel app.  
The government approved a draft law aligning Polish rules with eIDAS 2.0\. The wallet will be voluntary, start as a separate app reached through mObywatel and later merge with it; it will handle identity and age checks, qualifications and professional licences, cross-border services and qualified e-signatures, with selective disclosure, said digital minister Krzysztof Gawkowski. Eighty organisations are using its test environment.

[Spain’s Supreme Court demands security evidence for digital ID voting devices](https://www.biometricupdate.com/202611/spains-supreme-court-demands-security-evidence-for-digital-id-voting-devices) — *Biometric Update*  
Why it matters: Spain's Supreme Court asks for proof that 61,603 digital-ID devices for the election were security-checked.  
The court gave the Interior Ministry two days to confirm that the devices rented to verify the digital DNI (via a QR code from the MiDNI app) at polling stations for the 29 November general election passed the legally required security checks, with technical documentation. The request follows a challenge by the party Iustitia Europa to the use of MiDNI for voter identification, on which the court has not yet ruled.

[NIST gives relying parties a playbook for accepting mDLs online](https://www.biometricupdate.com/202611/nist-gives-relying-parties-a-playbook-for-accepting-mdls-online) — *Biometric Update*  
Why it matters: NIST's draft guide on accepting mobile driving licences online is open for comment.  
The second public draft of SP 1800-42 from NIST's NCCoE gives relying parties a shared threat model, privacy method and usability lessons for accepting mDLs over the internet, with a volume for banks' customer-identification rules; comments are open until 20 December. The same ISO standards underpin EU wallet credentials.

---

## Health & Cyber

**\[P1\]** [Access Dental Management Breach Affects Over 341,000 Individuals](https://www.hipaajournal.com/access-dental-valley-family-health-care-ochsner-health-data-breach/) — *The HIPAA Journal*  
Why it matters: A cyber outage diverts ambulances from a New Hampshire hospital; Access Dental breach hits 341,000.  
Monadnock Community Hospital, a critical access hospital, has had network and phone outages since 9 November: elective surgeries were cancelled for patient safety and ambulances remain on diversion, with radiology and lab services down until at least 23 November. Separately, Access Dental Management reported a breach of names, Social Security numbers and health records for 341,039 Texans, claimed by Worldleaks in June.  
severity high · EU: GDPR Art. 9, NIS2, EHDS · actor Worldleaks (Access Dental claim) (50%)

**\[P3\]** [Healthcare Sector Disclosed 100 Ransomware Attacks in Q3](https://www.hipaajournal.com/q3-2026-ransomware-report-healthcare/) — *The HIPAA Journal*  
Why it matters: Healthcare leads disclosed ransomware attacks in Q3, with 100 of 353, BlackFog counts.  
BlackFog's Q3 2026 report counts 353 disclosed and 2,668 undisclosed ransomware attacks, up 20% and 77% on a year earlier; healthcare had 100 disclosed and 239 claimed on leak sites, 98% of disclosed attacks involved data theft and the average demand was $763,253.  
severity medium · exploited in the wild · EU: NIS2, GDPR Art. 9

[Healthcare firms fail to monitor AI tools as shadow AI leads to breaches](https://www.healthcaredive.com/news/healthcare-ai-cybersecurity-monitoring/875171/) — *Healthcare Dive - Latest News*  
Why it matters: Hospitals rely on AI and SaaS vendors to tell them about breaches, a survey finds.  
Six in ten healthcare respondents to a DNSFilter survey rely on their AI and SaaS vendors to alert them to incidents, against 43% in other sectors, and under a quarter count on staff to report AI-related incidents; the sample is small (30 healthcare IT and security professionals), Healthcare Dive reports.

---

## Defence & National Security

[SpaceX's Starshield troubles in Ukraine could threaten its broader military business](https://arstechnica.com/space/2026/11/incoming-house-chair-threatens-spacex-funding-over-ukraine-controversy/) — *Ars Technica*  
Why it matters: The incoming House intelligence chair threatens SpaceX's government contracts over Starshield in Ukraine.  
Rep. Jim Himes wrote to Elon Musk that SpaceX's possibly unilateral change to a government contract, and its lack of response to Congress, call into question every federal contract with the company, and sent a classified message to DNI Jay Clayton saying funding systems with SpaceX as prime vendor would be hard to authorise. The inquiry concerns interruptions to Starshield service in parts of Ukraine, Ars Technica reports.

[Army plans to work with OpenAI on Project Gatekeeper](https://defensescoop.com/2026/11/20/army-project-gatekeeper-openai-gabe-chiulli/) — *DefenseScoop*  
Why it matters: The US Army will work with OpenAI to automate software accreditation for its networks.  
Project Gatekeeper lets vendors submit code to an Army DevSecOps platform for automated checks instead of the months-long authority-to-operate process, then moves approved code into an IL5 environment, acting Army CIO Gabe Chiulli said; OpenAI is among the first partners and the Army seeks reciprocity with the other services.

---

## Quantum & Cryptography

[MiniMax Reports Cryptanalysis Results From an Unreleased Internal Model](https://pandaily.com/minimax-mim-cryptanalysis-research-internal-model-disclosure) — *Pandaily - China Tech News, AI & Electric Vehicle Insights*  
Why it matters: MiniMax reports automated cryptanalysis results from an unreleased internal model.  
MiniMax's research blog describes experiments with an internal system it calls MIM on past NIST competition candidates, peer-reviewed cryptographic constructions and mathematical challenge problems, Pandaily reports. The company says the work does not affect deployed standards; past candidates are not the standards eventually chosen, and the results are self-reported. Excerpt-based.

[A game you cannot win unless you have a quantum computer](https://research.google/blog/a-game-you-cannot-win-unless-you-have-a-quantum-computer/) — *The latest research from Google*  
Why it matters: Google measures quantum contextuality as a benchmark on up to 105 qubits.  
Google Quantum AI researchers turned contextuality, the property that lets quantum players always win a coordination game no classical strategy can win every round, into something they measure and benchmark on a superconducting processor at up to 105 qubits. It is a test of genuinely quantum behaviour at scale rather than a cryptographic threat.

---

## Cybersecurity & Threats

**\[P2\]** [Bell Canada says 'agentic' hacker stole enterprise customer data](https://www.bleepingcomputer.com/news/security/bell-canada-says-agentic-hacker-stole-enterprise-customer-data/) — *BleepingComputer*  
Why it matters: Bell Canada says an "agentic" attacker stole enterprise customer data from its circuit-ordering system.  
Bell confirmed that on 17 November an agentic threat actor accessed several systems, including a pre-production environment of its Maximo system used to order and provision telecom circuits, and took enterprise customers' business contacts, ticket titles and some IP addresses of network elements; Bell says credentials and attachments were not in the dataset.  
severity high · exploited in the wild · EU: NIS2, GDPR

**\[P2\]** [U.S. Department of Agriculture grapples with security breach as IT staff take multiple systems offline](https://databreaches.net/2026/11/20/u-s-department-of-agriculture-grapples-with-security-breach-as-it-staff-take-multiple-systems-offline/) — *DataBreaches.Net*  
Why it matters: A cyber incident takes USDA payment tools for farmers offline for days.  
Nextgov/FCW and Government Executive report that numerous Department of Agriculture applications, including financial tools for payments to farmers such as crop insurance and disaster aid, have been offline for several days after an internally disclosed incident; the attacker and any data theft are unknown.  
severity high · EU: NIS2

**\[P1\]** [F5 releases emergency patches for critical BIG-IP RCE flaws](https://www.bleepingcomputer.com/news/security/f5-releases-emergency-patches-for-critical-big-ip-rce-flaws/) — *BleepingComputer*  
Why it matters: F5 ships emergency hotfixes for five critical BIG-IP flaws with no workaround.  
Five stack-based buffer overflows in BIG-IP APM and iRules (CVE-2026-121805, -120292, -120285, -116953, -123637) let unauthenticated attackers crash TMM and run code on vulnerable configurations; a sixth high-severity APM SAML flaw causes denial of service. F5 offers no mitigation; Shadowserver sees about 14,600 IPs with APM fingerprints online.  
severity critical · `CVE-2026-121805` · EU: NIS2, CRA

**\[P1\]** [Hackers Exploit Critical Cisco Expressway Flaw, Breach Four Devices and Reach Internal Network](https://thehackernews.com/2026/11/hackers-exploit-critical-cisco.html) — *The Hacker News*  
Why it matters: A public exploit for Cisco Expressway is used to breach a Russian organisation's internal network.  
F6 reports the first in-the-wild exploitation of CVE-2026-76466 in Cisco Expressway: attackers took two internet-facing devices as root and reached two internal cluster nodes, then stole credentials with scripts possibly written with AI help. CISA added the flaw to its KEV catalogue this week; patches came on 21 October.  
severity critical (CVSS 9.8) · exploited in the wild · `CVE-2026-76466` · EU: NIS2

**\[P2\]** [Cyberattack on service provider exposes payee data at 40 Irish credit unions](https://www.theregister.com/security/2026/11/19/cyberattack-on-service-provider-exposes-payee-data-at-40-irish-credit-unions/5310979) — *www.theregister.com - Articles*  
Why it matters: Ireland's minister faults Wellington IT for weeks of delay after a breach at 40 credit unions.  
Attackers exploited a flaw in Wellington IT's online banking platform between 2 and 4 October, exposing saved payees (names, IBANs, BICs) linked to members at about 40 Irish credit unions; the flaw was closed on 6 October but the credit unions learned of it only this month, and minister Robert Troy called the delay "not acceptable".  
severity high · EU: DORA, NIS2, GDPR