Good morning from the wAIve team in London.

Today, Tuesday 3 November 2026: A day for warnings and second thoughts. The National Cyber Security Centre now expects AI-powered ransomware to hit Britain's power stations and hospitals by 2028, and Goldman Sachs has taken its AI coding tools away from the traders and bankers it was cheering on in July. Meanwhile, Anthropic grew its quarterly revenue twelvefold and still managed to lose $23 billion doing it.

  POLICY  

The National Cyber Security Centre has issued a formal assessment warning that AI-powered malicious software is "highly likely" to be used against UK critical national infrastructure by 2028. The NCSC says AI will probably enable a ransomware attack against facilities spanning energy, transport, health, defence, and finance within that window.

THE PLAY: This is not a theoretical projection. It's an intelligence assessment from the state body responsible for protecting British digital infrastructure. Check your cyber insurance now.

  INDUSTRY  

Goldman Sachs has pulled AI coding tools, including Claude Code and GitHub Copilot, from its traders and investment bankers, restricting access to engineers only. As recently as July, Goldman's chief information officer Marco Argenti praised "vibe coding" by non-technical staff. The bank now cites productivity and risk management as its reasons for pulling back.

THE PLAY: This is one of the clearest signals yet that the initial wave of unrestricted AI tool rollouts inside major financial institutions is meeting resistance from compliance and ops teams. Expect other banks to follow.

  BREAKING  

Web hosting company Squarespace has announced it is cutting 19% of its workforce, around 300 people, as part of a broader restructuring. Most of the cuts fall on product and engineering. The company says the changes are meant to simplify how it operates and has not linked them to AI, although they come as AI tools reshape how websites get built.

THE PLAY: Software co's are now asking whether AI makes their human workforce redundant in the same way they once made developers redundant for their customers.

New figures disclosed ahead of Anthropic's IPO show the company grew quarterly revenue twelvefold to almost $17 billion but posted a $23 billion net loss in the process. Most of that loss is an accounting charge on the convertible notes it used to raise money. The business itself lost $567 million, having made a profit the quarter before.

Why it matters: A 12x revenue increase shows momentum. But even without the accounting charges, Anthropic slipped back into an operating loss as compute costs climbed. Investors buying into the IPO are making a big bet.

Saudi Aramco has agreed to acquire a 49% stake in Humain, the Saudi government-backed company established to turn the kingdom into a global AI hub. The $2.6 billion deal brings the world's largest oil company directly into the AI infrastructure race and gives Humain access to Aramco's capital and energy resources.

The regional angle: For European and UK businesses operating in the Gulf, this deal signals that Saudi Arabia is accelerating its AI ambitions with serious industrial backing, not just sovereign wealth.

Behavioural AI startup 42AI has emerged from stealth with $62.5 million in funding, backed by a world model it claims can predict human behaviour in novel situations before a person has decided to act. Ten brokerages already use it, and the company plans to move into gaming, fraud prevention and cybersecurity.

Why it matters: Predictive behavioural AI raises immediate questions about consent, regulation, and liability. If 42AI's claims hold under scrutiny, the technology has significant commercial potential.

Closing prices, Monday 2 November (US markets). Not investment advice.

A research assistant that searches the web in real time, synthesises results from multiple sources, and cites every claim with a link.

THE PLAY: Ask it to summarise what has happened in the last 30 days on a topic that matters to your business. Review the sources it cites and forward the summary to your team as a briefing note.

Google's AI research tool that reads documents, reports, and articles you upload, then answers questions, creates summaries, and generates audio briefings from your own materials.

THE PLAY: Ask NotebookLM to identify the five most significant risks and the three decisions it implies you need to make. Use the output as your meeting prep.

For three decades, websites were built for human eyes. Every design decision, every sponsored listing, every subscription paywall assumed a person was sitting at the other end. Until now.

AI agents are now browsing, shopping, booking, and researching. The companies that built the commercial internet around human attention are discovering that their model no longer works for users that can't be distracted with a click on sponsored results. MIT Technology Review's Grace Huckins calls this the agentic web, and it is arriving faster than most businesses anticipated.

When AI summaries appear in search results, users no longer need to click through to the underlying article. One analysis found that publisher revenue from certain advertising formats fell by roughly a third between 2025 and 2026 as AI-generated overviews reduced click-through rates.

E-commerce faces a parallel challenge. Research from the University of Toronto found that AI agents are less likely to choose products tagged as "sponsored" in search results, undermining one of the core revenue routes for online retail. Meanwhile, Amazon blocked Meta's Muse agent from its site entirely, preferring to control what its own shopping agent recommends rather than cede that influence to a competitor.

The mobile web promised liberation from desktop platforms and delivered instead a duopoly of app stores. The agentic web carries the same risk.

A business that doesn't appear favourably in GEO (Generative Engine Optimisation) may be as invisible as one that never ranked on Google in 2010.

Takeaway: First, audit your agent-readiness. Can an AI agent find your products, services, and pricing through a structured query? Many business websites are optimised for human navigation, not agent searches.

Second, if your commercial model relies on human clicks, dwell time, or response to visual advertising, you need a plan for a world where your customer isn't clicking.

The businesses that act now on agent-readiness will have a structural advantage over those that wait for the disruption to become undeniable.

Most disappointing AI answers come from thin briefs. Rather than trying to write the perfect prompt, tell the model what you want and ask it to question you before it starts. It will spot the gaps in your brief far faster than you will, and the answer that follows is built on your context rather than its guesses.

This works because the model stops guessing about your audience, your constraints and what good looks like. Five minutes of answering questions usually saves three rounds of rewrites.

Thanks for reading, the wAIve Daily Team. Found this useful? Forward it to someone who needs it. Tell us what you need to know. Reply with your questions.

wAIve Daily is published with Centai Strategy