Good morning from the wAIve team in London.

Today, Tuesday 20 October 2026: Remember that friendly White House meeting between Presidents Trump and Xi? Washington has now banned DeepSeek from government systems, reviving the old China tension. Elsewhere, Anthropic opens $45 billion in deal talks with UK firm Nscale, and a single selfie can now predict cancer survival odds. Happy reading.

  POLICY  

The Department of Homeland Security has ordered that DeepSeek should be from civilian government systems. The order was made public this week but has applied since the end of July, and covers agencies such as the Treasury, State and Justice departments, though the Department of Defense is exempt. It follows months of Trump administration criticism of Chinese AI firms, and claims from Anthropic and OpenAI that Chinese developers have copied their models.

The play: If any of your vendors build on DeepSeek, directly or through a sub-supplier, check now whether that exposure touches public sector work.

  RESEARCH  

Doctors in Massachusetts have developed an AI tool that estimates a patient's biological age and predicts cancer survival outcomes from a single photograph of their face. The system was built to read biological age signals in facial features that correlate with clinical outcomes.

In related trials, more than eight in ten patients using an associated treatment system were able to read, use their phones, and complete everyday tasks after a training period.

The play: Regulators will take a close interest in how such tools are deployed. If you operate in healthcare, insurance, or occupational health, ask your legal/compliance team to check the regulatory status of AI-based facial analysis tools.

  BREAKING  

Microsoft has issued a formal warning to enterprise customers: AI agents that employees deploy using systems outside the control of major AI labs should be treated as potential insider threats. The company is calling for organisations to monitor these agents closely, apply the same security principles used for human users, and assume that an agent connecting to internal systems it was not explicitly designed for represents a risk.

Microsoft is not saying agents are malicious. It is saying their behaviour is sufficiently unpredictable that the only prudent posture is to treat them with the same scepticism applied to a new contractor with broad system access.

The play: This is a shift in how enterprises should think about agentic deployment governance. Organisations that have deployed AI agents across internal tools without a formal access review process are exposed.

Reports indicate that Anthropic is in advanced discussions with UK infrastructure firm Nscale for a compute partnership worth up to $45 billion. If confirmed, it would be the latest in a string of landmark infrastructure commitments Anthropic has made over the past year as it races to secure the processing power needed to train and run frontier AI models at scale.

Why it matters: A $45 billion compute agreement would be one of the largest single infrastructure commitments in the history of the technology industry. AI labs now require supply chains of a size previously associated only with national energy or defence programmes. The sums are off the scale, and it isn't ending anytime soon.

The Nasdaq 100 dropped five and a half percent last week in one of its sharpest falls in over three years.

The sell-off was driven in part by growing investor concern that the estimated one trillion dollars being committed globally to AI infrastructure in 2026 may not generate returns at the scale or speed originally projected. The International Monetary Fund raised the question of AI profitability in its latest assessment of the global economy.

Why it matters: The market is beginning to price in the gap between AI capability and demonstrated revenue. If you're using AI investment valuations to justify AI spend, revisit the numbers before the next board meeting!

Tenstorrent, the AI chip company founded by chip designer Jim Keller, has closed a $1.5 billion funding round, lifting its valuation to $6.2 billion. The raise comes after the company attracted acquisition interest earlier this year and chose instead to remain independent. Tenstorrent's approach centres on open-weight AI chips, designed to give enterprises an alternative to the Nvidia ecosystem.

THE PLAY: Ask your infrastructure team this week what percentage of your AI compute spend is currently tied to a single vendor. If the answer is above eighty percent, commission a brief assessment of what switching costs and alternatives look like.

A personal AI assistant built by Eugenia Kuyda, the founder of Replika, that helps users think through decisions, reflect on priorities, and navigate complex situations in daily life and work. Best for executives who want a private, low-friction thinking partner available outside working hours. Free to download on iOS and Android at wabi.app.

THE PLAY: Open Wabi and describe the situation in two minutes of voice note. Ask it to surface the decision you are actually avoiding. Compare its response against your own instinct before acting.

An AI companion app that helps users navigate decisions and situations across professional and personal life, with over 250,000 active users and $30 million in fresh funding this month. Best for leaders who want an always-available sounding board for low-stakes thinking before escalating to their team. Free tier available at tolan.ai.

THE PLAY: Describe the situation to Tolan and ask it to identify what the other person is most likely to be worried about. Use the response to prepare your approach rather than react in the moment.

Agentic web traffic tripled between August and September this year. Meta's Muse AI agent was downloaded five million times in its first month. OpenAI's Dots is doing similar work. These agents don't browse the web the way a person does. They operate at machine speed, hit the same pages repeatedly. The result is a growing standoff between agents and platforms built to serve humans.

Amazon has blocked Meta's Muse agents from its marketplace. It's now fighting a legal battle against Perplexity to prevent its agents from accessing Amazon's systems. Pizza Hut blocks agents. Yelp refuses to allow AI to make restaurant reservations. Platforms that spent years optimising their user experience for human attention are discovering that agents strip away the friction those experiences depend on, including the friction that generates ad revenue.

When an agent shops on your behalf, it ignores the ads. Shopify has moved in the opposite direction, building dedicated checkout lanes for AI agents.

Every organisation that sells through digital channels should now consider two questions.

It is already happening in trials, and the legal framework governing it does not yet exist.

The rules for how agents interact with websites are still being written. Commission a review this quarter of how AI agents currently interact with your digital platforms. Design for an agent-first internet now and set the terms ahead of time.

When an AI model gives you a recommendation, it is optimised to sound confident. That confidence can mask weak reasoning or suppressed alternatives. The self-consistency check forces the model to stress-test its own answer before you act on it.

Apply this after any AI-generated strategic recommendation, vendor comparison, or risk assessment. The technique works because it removes the model's default tendency to defend what it has already produced, forcing it to surface the dissenting logic it suppressed the first time around.

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