> ## Content Index
> Fetch the complete content index at: https://longtalkshort.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# The Wave and the Narrow Path: DeepMind's Co-Founder on Why AI Can't Be Stopped, Only Contained
- URL: https://longtalkshort.com/the-coming-wave-mustafa-suleyman-summary-containment/
- Published: 2026-10-28T16:38:10.000Z
- Updated: 2026-10-28T16:38:10.000Z
- Description: Mustafa Suleyman's The Coming Wave argues AI and synthetic biology can't be stopped, only contained, along a narrow path between catastrophe and dystopia. His four features, ten steps, and whether the plan holds.
- Author: Truman
- Tags: Science, Mustafa Suleyman

**One of the people who built modern AI opens his book with a chapter called "Containment Is Not Possible."**

Two chapters from the end comes the counter-title: **"Containment Must Be Possible."** Mustafa Suleyman's *The Coming Wave* lives in the gap between the two.

Suleyman co-founded DeepMind, the lab Google bought in 2014 and whose AlphaGo beat the world's best Go players. He later co-founded Inflection AI and, since 2024, has run Microsoft AI. So the book (2023, written with Michael Bhaskar) reads as **an insider's confession.**

*In short: AI and synthetic biology form one wave of general-purpose technology that is cheap, fast-improving and impossible to un-invent. Suleyman argues we must contain it anyway, steering between catastrophe and a surveillance dystopia, and offers ten steps. The diagnosis is strong. The plan relies on cooperation that hasn't shown up.*

## 1\. Every wave proliferates

Suleyman reads history as a series of waves: fire, writing, the printing press, the steam engine, electricity, the internet. Each general-purpose technology **gets cheaper, spreads and gets built on,** whatever its inventors intended. Bans rarely hold. **Proliferation, he says, is technology's default state.**

The coming wave has two cores:

- **AI** turns data into capabilities: seeing, writing, coding, planning.
- **Synthetic biology** turns DNA into something you can read, edit and print, as sequencing and synthesis costs collapse.

Each speeds up the other, and both spill into robotics, quantum computing and energy. His yardstick for AI is a **"Modern Turing Test":** can an AI, given $100,000, turn it into $1 million by researching, designing, making and selling a product with little human help? He described the test in a 2023 *MIT Technology Review* interview.

So why is this wave harder to hold back than the last ones?

## 2\. Four features that make it hard to contain

![Four cards: asymmetry, hyper-evolution, omni-use and autonomy, with a bar beneath for pessimism aversion](https://storage.ghost.io/c/d7/93/d793216c-cf13-441f-bdae-4f3aef998c0e/content/images/2026/10/fig1-32.png)

The four features come from the book; the one-line glosses are ours. Made for Long Talk Short.

- **Asymmetry:** tiny actors can have outsized impact. Think of a cheap drone against an expensive tank, or one person with a DNA synthesizer.
- **Hyper-evolution:** tools improve on a cycle of months, faster than law and institutions adapt.
- **Omni-use:** the model that designs a drug can, with different prompts, help design a toxin.
- **Autonomy:** systems increasingly act without a human approving each step.

He adds a human weakness: **"pessimism aversion,"** the habit of elites and everyone else to wave away dark scenarios because they're uncomfortable. Add the incentives (national rivalry, profit, prestige, curiosity) and you get the containment problem: **how do we keep power over tools this useful, this cheap and this easy to copy?**

## 3\. Two ditches and a narrow path

Both easy answers fail.

- **Let the wave run open** and you risk catastrophe: engineered pandemics, cyberattacks on infrastructure, floods of disinformation weakening fragile states.
- **Lock it down hard** and you risk dystopia: governments with total surveillance and the power to block any lab, any line of code, any garage.

The nation-state is caught both ways, **empowered by the same technology that is eroding its authority.** The only route is a "narrow path" between the two. How do you walk it?

## 4\. The ten steps

The final chapter lays out ten layers, from the technical core outward:

- **1\. Safety:** a major research push on AI and bio safety.
- **2\. Audits** that verify safeguards actually work.
- **3\. Choke points,** such as advanced chips and their supply chains, to slow the pace.
- **4\. Makers:** critics who build, not just comment.
- **5\. Businesses** whose structure ties profit to purpose.
- **6\. Governments** that build technical skill, license frontier work and regulate.
- **7\. Alliances:** a treaty on the scale of nuclear arms control.
- **8\. Culture** that shares failures openly, as aviation learns from crashes.
- **9\. Movements** of citizens demanding all this.
- **10\. Coherence,** so the layers reinforce one another instead of leaving gaps.

> It's time for an Apollo program on AI safety and biosafety.

And on step four, in his words: "credible critics must become practitioners."

![Ten cards for the ten containment steps, each with a short note on its status in 2026](https://storage.ghost.io/c/d7/93/d793216c-cf13-441f-bdae-4f3aef998c0e/content/images/2026/10/fig2-26.png)

The ten steps are the book's. The 2026 status notes are our rough read, not a formal assessment. Made for Long Talk Short.

## Weighing it: what holds up

We read two prominent reviews, by John Naughton in *The Guardian* and by Carl Mitcham and Lukas Fuchs in *Issues in Science and Technology*, and checked the policy record since publication.

**The diagnosis holds.** Proliferation, dual use and racing incentives aren't controversial, and the book is unusually candid for a lab founder; Naughton calls its handling of uncomfortable questions "the most refreshing feature of the book." Some steps have partly happened since 2023: US export controls on advanced AI chips, the EU AI Act in 2024, and a Council of Europe framework convention on AI opened for signature in 2024 and signed by the US, the UK and the EU.

**The plan is where critics push back.** Naughton calls the solution "effectively a utopian dream," while praising Suleyman for pairing pessimism of the intellect with optimism of the will. Mitcham and Fuchs find many proposals "deeply naive and unrealistic": Apollo had one clear goal, but AI safety is messier with no finish line. They ask whether trusting the EU or US to lead global rules is itself pessimism aversion. Historians of technology find the wave metaphor too simple, and Naughton notes the book sometimes sounds like the technological determinism it disclaims.

**Two tensions of our own.** The plan leans on the same companies and racing states it diagnoses as the problem, and its author now runs a frontier AI lab at one of them. That's not hypocrisy, since step four is his own, but it means success depends on incentives he admits push the other way. And the book offers no clear test for when the narrow path has been lost, which makes the plan hard to judge as a forecast.

**Related reading:** Microsoft chief executive Satya Nadella makes his own case for containing models in [Own the Test, Rent the Model](https://longtalkshort.com/satya-nadella-own-your-ai-eval-open-vs-closed-models/); for what AI is already doing to mathematics, see [The Proof Came First](https://longtalkshort.com/ai-math-proofs-navier-stokes-riemann-lex-fridman-503/); and for the money behind the race, [AI Isn't the Bubble. The Debt Behind It Might Be.](https://longtalkshort.com/ai-bubble-first-domino-debt-grand-angle/)

## Long Talk Short

- Technology's default is proliferation: general-purpose tools get cheaper, spread and get built on, and bans rarely hold.
- The coming wave is AI plus synthetic biology, harder to contain because of asymmetry, hyper-evolution, omni-use and autonomy.
- "Pessimism aversion" keeps leaders from facing the dark scenarios.
- The goal is a narrow path between open catastrophe and locked-down dystopia.
- His ten steps run from safety research and audits through chip choke points, governments and treaties to culture and movements, held together by coherence.

**The one thing to remember:** judge AI governance by whether its layers actually reinforce one another. The book's diagnosis is strong; the cooperation its cure needs is still mostly missing.

Source: *The Coming Wave: Technology, Power, and the Twenty-First Century's Greatest Dilemma* by Mustafa Suleyman with Michael Bhaskar (Crown, 2023). This is an independent summary and is not affiliated with the authors or publisher.

*This article is for general information only and is not legal, policy or investment advice.*