Your Weekly AI Briefing

Goldman Sachs has quietly barred all Hong Kong-based employees from accessing Anthropic's Claude, Bloomberg and the Financial Times reported on 29 April. The restriction followed a strict internal review of the bank's contractual arrangements with Anthropic, in consultation with Anthropic itself, which concluded that Hong Kong falls outside Claude's supported markets. The block is geography-specific and applies even to Goldman staff visiting Hong Kong from other offices. Other AI tools — including OpenAI's ChatGPT and Google's Gemini — remain available on the bank's internal platform. Anthropic confirmed that Claude has never been officially supported in Hong Kong.

Why it matters: Goldman may have done the legal due diligence that competitors have not. Every multinational financial institution with an enterprise AI contract needs to audit whether geographic licensing terms create compliance exposure — or an unequal AI capability across its global workforce. The next question is whether other major banks follow.

The Wall Street Journal reported on 28 April that OpenAI fell short of several monthly internal revenue targets in early 2026 and failed to reach its goal of 1 billion weekly active users by year-end 2025. Market share losses to Anthropic in coding and enterprise, and to Google's Gemini more broadly, were cited as the primary cause. CFO Sarah Friar privately warned colleagues that OpenAI may struggle to fund future data-centre commitments if revenue growth does not accelerate — a pointed concern given the company's $250 billion Azure compute commitment and a planned IPO at a valuation of approximately $1 trillion. OpenAI and Altman called the report "ridiculous" in a joint statement. Oracle shares fell 4%, Nvidia and Broadcom dropped 3–4%, and CoreWeave fell over 5% on the news.

Why it matters: A company with a $250 billion compute commitment and a planned IPO cannot afford sustained revenue underperformance. If the gap between OpenAI's infrastructure spend and its revenue trajectory widens, pricing pressure on enterprise contracts, product roadmap changes, or IPO delays become live risks for every organisation that has built its AI strategy around OpenAI products.

Why it matters: AI's power problem is now a capital allocation problem. Meta is committing balance sheet to energy technologies that don't yet exist at commercial scale — a signal that the hyperscalers have concluded the grid cannot deliver what their AI roadmaps require. Competitors facing the same constraint will be forced to respond with similar bets or accept an infrastructure disadvantage.

The White House told Anthropic on 30 April that it opposes the company's plan to expand access to Mythos — its advanced AI model capable of autonomously finding and exploiting software vulnerabilities — from approximately 50 organisations to 120, according to the Wall Street Journal, confirmed by Bloomberg. Mythos, launched in early April under Project Glasswing, was initially made available only to a small group including Amazon, Google, and JPMorgan. The administration cited two concerns: misuse risk and insufficient compute capacity to serve both government agencies and an expanded user base. The NSA is among agencies currently testing Mythos. The Pentagon separately maintains Anthropic is a national security supply chain risk, following a breakdown over autonomous weapons use. Defence Secretary Pete Hegseth called Anthropic's CEO an "ideological lunatic" during Congressional testimony on 1 May.

Why it matters: This is the first time the US government has formally intervened to block an AI company from expanding access to its own commercial product. Every enterprise using Anthropic models should monitor this closely: a company simultaneously barred by the Pentagon, blocked from expanding its most advanced model, and subject to a possible executive order represents an unusual concentration of regulatory risk for any procurement team to absorb.

Courts in Hangzhou ruled on 30 April that a technology company's dismissal of a senior employee — whose role had been automated by AI and who was offered a replacement position at a salary cut from 25,000 to 15,000 yuan — constituted illegal termination. The Hangzhou Intermediate People's Court held that AI adoption is a controllable business strategy, not an unavoidable economic disruption, and that the cost of technological transformation cannot be transferred to the employee. The case is among several emerging from Chinese cities including Beijing, where a data-mapping worker replaced by AI won through arbitration last year on the same principle. Both rulings were timed to coincide with China's May 1 Workers' Day.

Why it matters: Any multinational manufacturer or technology company with significant operations in China now faces legal exposure if AI-driven restructuring results in dismissals without retraining or fair redeployment. The ruling creates a direct cost to automation in the world's largest manufacturing economy — and signals a legislative direction that other jurisdictions may follow.