#podcast

Weekly Security Roundup — week ending 22 November 2026

China-linked AI agents autonomously exploit zero-day; FBI NCIC breached twice in two months; UK admits Apple backdoor demand; JPMorgan/Visa/Mastercard vendor breach; AI supply chain leaking data at scale.

🎙 Apple Podcasts 🎙 Spotify 📡 RSS Feed
audio-thumbnail
Weekly Security Roundup — week ending 22 November 2026
0:00
/10:44

Security Stories

China-Linked AI Agents Autonomously Exploit MONAI Label Zero-Day

Amazon researchers confirmed a China-linked threat actor used the open-source Cairn AI penetration testing framework to autonomously discover and exploit a critical zero-day in MONAI Label — a medical imaging AI tool — then used that foothold to compromise university servers. This is the first confirmed real-world case of AI agents conducting the entire exploitation kill chain without human-written exploit code. (SecurityWeek, BleepingComputer, The Record)

FBI NCIC Hacked — Second FBI Breach in Two Months

The FBI's National Crime Information Center — the central law enforcement database queried by every US agency hundreds of millions of times per year — was breached, following an earlier FBI system compromise in September. The FBI has not disclosed the attack vector, duration, or scope of data accessed. The NCIC holds records on wanted persons, witnesses in protection programmes, and sensitive law enforcement intelligence. (SecurityWeek, BleepingComputer, BBC)

UK Government Forced to Admit Apple Backdoor Demand

The Investigatory Powers Tribunal rejected the Home Office's argument that it couldn't even acknowledge issuing a secret Technical Capability Notice to Apple — effectively ordering the company to build backdoor access into iCloud's end-to-end encrypted storage. Apple withdrew Advanced Data Protection from UK users in February. The IPT's ruling strips the government of deniability; full details of what was demanded remain sealed. (The Register, Slashdot)

JPMorgan, Visa, Mastercard: N0n Group Exploits Vendor Breach

JPMorgan Chase confirmed a third-party supplier breach exposed customer personal and marketing data, following claims by the N0n extortion group that it had stolen data touching Chase, Visa, and Mastercard customers. Stolen records include names, emails, phone numbers, and purchase history — sufficient for targeted phishing but not direct financial fraud. The incident highlights cascading risk from shared marketing and analytics vendors. (SecurityWeek, BleepingComputer)

N-able Breach Causes MSP Outage; 40+ Irish Credit Unions Hit by Supplier Compromise

N-able disclosed a cyberattack caused a major outage of its Mail Assure and SpamExperts cloud email filtering services, affecting thousands of managed service providers globally. Separately, more than 40 Irish credit unions discovered a shared Belfast-based data processor had been breached — with the compromise going undetected for six weeks before the Irish Data Protection Commission received over 40 simultaneous breach notifications. (BleepingComputer, The Record)

AI Supply Chain Leaking User Data at Scale

A researcher planted 2 million canary tokens across AI service infrastructure and found user prompts leaking through multi-provider inference chains at an alarming rate. OpenRouter and Together.ai both disclosed data processing incidents this week, with inference providers handling user data beyond their agreed terms. Bell Canada described its own attacker as "agentic" — using AI-driven autonomous techniques to breach its telecom provisioning platform. (Ars Technica, BleepingComputer)

Microsoft Entra Connect Health Exploited; Rackspace Spot Breach

Microsoft confirmed its Entra Connect Health hybrid identity monitoring service had been actively exploited in the wild, with attackers stealing tenant configuration data and administrator credentials. Separately, Rackspace disclosed its Spot cloud cost optimisation platform was breached via compromised credentials, with attackers accessing Kubernetes secrets in customer clusters. Both incidents underscore the risk of cloud management and monitoring tooling as an attack vector. (BleepingComputer, BleepingComputer)

BADBOX 3.0 — 40 Billion Fraudulent Ad Requests Per Day

The BADBOX Android device botnet returned in its third generation, generating a record 40 billion daily fraudulent ad requests. The botnet targets consumer devices with supply-chain-implanted backdoors and also sells proxy and remote access capability on criminal markets. Disrupted before and rebuilt before, BADBOX's persistence reflects unresolved security incentive problems in low-cost device manufacturing. (BleepingComputer)

Signal Blocks AI Agents; CTO of RansomwareHelp Indicted; White House Cyber Leadership Reshuffled

Signal updated its platform to block Google Gemini, Apple Siri, and other AI agents from accessing messages without explicit user permission. The CTO of ransomwarehelp.com — a Florida company that claimed to unlock ransomware-encrypted data — was indicted for wire fraud, having allegedly simply paid ransom while billing clients for a proprietary decryption service. And the White House replaced National Cyber Director Sean Cairncross with Pentagon CIO Kirsten Davies as Trump reshuffled federal cyber leadership. (Signal, Krebs, The Record)


Key Themes

AI as Offensive Weapon

The Cairn zero-day discovery, the EU Commission AI-assisted attack, and Bell Canada's "agentic" attacker together mark a qualitative shift in the threat landscape. AI is no longer purely a defensive tool or a productivity accelerator for defenders — it is being actively deployed to find vulnerabilities, automate exploitation, and conduct multi-step intrusions with minimal human direction. The debate about whether this represents a paradigm shift or simply faster human-equivalent attacks has real implications for vulnerability disclosure timelines and patch SLAs.

The Supply Chain Trust Crisis Deepens

Three separate supply chain incidents this week — the N0n vendor breach affecting JPMorgan/Visa/Mastercard customers, the N-able MSP outage, and the Belfast data processor exposure hitting 40 credit unions — reinforce that strong internal security controls offer diminishing returns if third-party risk isn't managed with equal rigour. The six-week undetected exposure in the credit union case is particularly troubling. Organisations need better continuous monitoring of their data processor ecosystems, not just point-in-time assessments.

Encryption Under Political Pressure

The UK Apple backdoor disclosure sits alongside the broader AI data leakage story to paint a picture of encryption and privacy under sustained attack from multiple directions simultaneously — state actors using legal mechanisms, and market actors using lax data handling. Signal's decision to explicitly block AI agents from accessing its message store is a noteworthy defensive response, but it also signals how normalised the assumption of AI surveillance has become.


NCSC / CISA

No new NCSC blog posts were published this week. CISA added no new Known Exploited Vulnerabilities to the KEV catalog this week. The most recent substantive NCSC guidance on AI risk — the August 2026 post on managing the cyber risk of agentic AI — remains highly relevant given this week's events.


HaveIBeenPwned Activity

BreachDate AddedAccountsData Types
Too Lost (music distribution platform)20 Nov 2026724,323Passport numbers, driver's licences, partial credit card data, payment histories, DOB, physical addresses
Careers360 (Indian education platform)20 Nov 20263,609,421Academic records, castes, email addresses, names, phone numbers