Today's brief covers stories from 19 to 20 November.
- EdgeA remote support appliance is compromised through a zero-day
- FootholdAttackers move to an internal appliance and gain root
- Control planeThe vCenter authentication bypass gives access to virtualisation
- PersistMalware hijacks tokens and creates administrator accounts
- ExfiltrateDatabases and virtual machine data are taken
- FixPatches arrive more than two months after exploitation began
1. The European Commission confirms an AI-assisted cyberattack
The European Commission confirmed that an AI-assisted cyberattack, detected on 22 October, disrupted staff services including its HR platform Sysper, the MyPMO medical reimbursement system and translation services, with some outages lasting more than ten days1. The Commission says it has found no evidence of a data breach so far and that the investigation continues; press reports link the campaign to a publicly available AI penetration testing framework, which the Commission has not confirmed2.
CTO take: the practical change from AI-assisted attacks is speed and persistence, not new techniques. Attackers can try more paths, faster, against everything you expose. That makes the basics more urgent: phishing-resistant sign-in, quick patching of internet-facing systems, segmentation so one compromised service does not reach the rest, and detection tuned to spot rapid, automated reconnaissance. European public bodies and their suppliers should expect questions on exactly these controls. See the AI cyber risk controls case study and my Europe page.
2. The vCenter zero-day campaign began with a BeyondTrust flaw
Google's Mandiant and Threat Intelligence Group attribute exploitation of the vCenter authentication bypass to Armory Castle, a suspected China-nexus espionage group, which first compromised BeyondTrust Remote Support instances through a separate zero-day in late August, then moved to an internal appliance and on to vCenter from early September3. Google says a small number of organisations are confirmed compromised and others are likely affected4.
CTO take: this is the pattern from this month's vendor admin access analysis playing out: a remote access tool at the edge, then the virtualisation control plane. If you run BeyondTrust Remote Support or vCenter, patch both, then hunt rather than assume: look for new administrator and single sign-on accounts, unusual vCenter logins and unexpected outbound traffic from management appliances since late August. Keep management interfaces on networks that remote support tools cannot reach directly.
3. Transport for London discloses an Oyster photocard breach
Transport for London said about 265,000 Oyster photocard customers were affected after it detected suspicious activity on 16 November and took the photocard system offline; reported exposed data includes names, contact details, addresses and security questions and answers5. TfL says passwords and payment card details were not compromised, and customers will be asked to reset passwords and security answers when the portal returns6.
CTO take: security questions are the weak point here. Once their answers leak, they are useful to attackers on every other service that uses the same questions, and customers cannot change their mother's maiden name. If your product still uses knowledge-based questions for recovery, replace them with stronger recovery factors. UK organisations should also note how quickly the regulator was informed. See the access hardening case study and my UK page.
4. A filing names Anthropic as tenant of a $28 billion Texas lease
Riot Platforms had disclosed a 15-year lease of its Corsicana, Texas campus to a frontier AI lab, with total contract value of about $28 billion7. Its latest quarterly filing names Anthropic as the tenant8.
CTO take: long leases like this are how AI providers secure future capacity, and they are a useful signal for customers. A provider that has locked in power and space is better placed to honour capacity commitments, but also carries fixed costs that shape its pricing. When negotiating, ask providers about capacity guarantees and regions, not just per-token prices. See AI capacity contracts and the multi-region AI capacity case study.
What I'd do this week
- Patch BeyondTrust Remote Support and vCenter, then hunt for new admin and SSO accounts since late August.
- Check that management interfaces are unreachable from remote support and user networks.
- Replace security questions in account recovery with stronger factors.
- Review detection for fast, automated reconnaissance against internet-facing services.
- Ask AI providers about capacity guarantees in your regions at the next renewal.
I am a fractional CTO and AI architect based in India, the United States and the UAE, working globally. If any of these lands on your desk this week, analyse your challenge and bring the result to a call.
- 1European Commission Confirms AI-Assisted Cyberattack Disrupted ServicesSecurityWeek · Nov 2026
- 2European Commission Confirms AI-Driven Cyber-Attack Took Systems OfflineInfosecurity Magazine · Nov 2026
- 3ARMORY CASTLE Exploits vSphereGoogle Threat Intelligence · 19 Nov 2026
- 4Chinese espionage group targets critical flaw in VMware vCenterCybersecurity Dive · Nov 2026
- 5Transport for London breach exposes 265,000 Oyster card users' dataBleepingComputer · Nov 2026
- 6265,000 Oyster photocard customers' data exposed in Transport for London breachHelp Net Security · 20 Nov 2026
- 7Riot Platforms, Inc. Form 8-K, Exhibit 99.1US SEC · 2026
- 8Riot Names Anthropic As Tenant In $28 Billion Texas LeaseBitcoin Magazine · Nov 2026