Today's brief covers stories from 18 to 19 November.
1. Broadcom patches a critical vCenter authentication bypass
Broadcom's advisory VMSA-2026-0009 fixes five flaws across vCenter, SDDC Manager and VMware Tools, the most serious being CVE-2026-123890, a 9.8-rated authentication bypass in the VMware Directory Service that lets an unauthenticated attacker with network access to vCenter gain access; there is no workaround1. BleepingComputer reports the flaw was exploited as a zero-day since September by China-linked attackers2; Broadcom's advisory itself is reported not to describe exploitation, so treat that as reported rather than confirmed by the vendor.
CTO take: vCenter is the control plane for every virtual machine it manages, so this is a patch-tonight item for anyone running it. Patch, then check: review vCenter authentication and administrative activity since September, and confirm vCenter is reachable only from a management network, never from user networks or the internet. If you are still on a version without a public fix, that is the moment to fund the upgrade. See admin tools as perimeter and the management plane hardening case study.
2. Malaysia tells data centres to follow US chip rules, as the US backs its sovereign AI cloud
The US rule on remote access to advanced semiconductors was published in the Federal Register on 19 November3. Malaysia's trade minister said data centre operators must comply, including licences for advanced chips, and that the government would notify approved data centre projects and seek clarification from the US on licensing and third-party services4. Separately, Bloomberg reports the US will offer $50 million to support Malaysia's sovereign AI cloud5.
CTO take: for anyone buying GPU capacity in Malaysia, the practical effect is more know-your-customer questions from providers and some uncertainty about how third-party cloud services will be licensed. Keep your ownership and end-use information ready, avoid single-provider dependence in the region while the detail settles, and check contract terms for what happens if a provider loses a licence. See the Malaysia GPU cloud case study and my Southeast Asia page.
3. UK financial regulators publish AI cyber risk practices
The Bank of England, PRA and FCA published a consultation on cyber risk and resilience management, CP17/26 and GC26/9, which clarifies how firms should meet existing requirements and is open until 26 February 20277. Alongside it, a separate effective practices paper sets out how firms and financial market infrastructures are managing cyber risk related to AI, including reducing their attack surface against AI-enabled attacks6.
CTO take: effective practices are not rules, but they become the benchmark supervisors use in conversations. UK banks, insurers and payment firms should map their AI use, including agents and third-party models, against the practices now, and respond to the consultation if the expectations do not fit how they actually operate. Firms outside the UK can use it as a free, well-reasoned checklist. See the UK cyber resilience case study and my UK page.
4. Anthropic's shares begin trading
Anthropic priced its IPO at $83 a share, raising more than $81 billion in the largest offering on record8. Its shares opened on Nasdaq on 19 November at $111.50, TechRepublic reports9, and later closed lower than that opening price.
CTO take: for customers, a listed AI supplier means more public disclosure about capacity, costs and risks, which is useful in vendor reviews, and more pressure on revenue, which can show up in pricing and packaging. Neither changes the basic advice: keep a model gateway so you can switch providers, and track each provider's terms as carefully as their benchmarks. See LLM model routing and the AI provider resilience case study.
What I'd do this week
- Patch vCenter, restrict access to a management network and review activity since September.
- If you use GPU capacity in Malaysia, ask providers how the US licensing affects your service and contract.
- UK financial firms: compare AI and agent use against the new effective practices and plan a consultation response.
- Check that you could move your main AI workloads to a second provider without a rewrite.
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.
- 1Critical VMware vCenter Zero-Day Flaw Lets Network Attackers Bypass AuthenticationThe Hacker News · Nov 2026
- 2VMware patches vCenter zero-day exploited in attacks since SeptemberBleepingComputer · Nov 2026
- 3Addressing Remote Access to Advanced SemiconductorsFederal Register · 19 Nov 2026
- 4Malaysia orders data centres to follow US rules on AI chipsCrypto Briefing · Nov 2026
- 5US, Malaysia to Cooperate on Sovereign AI Cloud With $50 Million SupportBloomberg · 19 Nov 2026
- 6Effective practices: Cyber risk and resilience management relating to artificial intelligence (AI)Bank of England · Nov 2026
- 7CP17/26 and GC26/9: Cyber risk and resilience managementBank of England · Nov 2026
- 8
- 9Anthropic IPO: Claude Maker Opens at $111.50 in Record Nasdaq DebutTechRepublic · Nov 2026