The Flood We Built With Our Own Hands

Every civilisation has a flood story. In the oldest versions, the waters rise not because of malice but because someone opened a gate that could not be closed. A sluice built with good intentions, overwhelmed by the volume it released. Mustafa Suleyman, who co-founded DeepMind and now leads Microsoft AI, argues that humanity has constructed precisely such a sluice. The twin currents of artificial intelligence and synthetic biology are pouring through it, carrying transformative promise and existential peril in the same torrent.

His book, The Coming Wave, is an attempt to map that torrent's trajectory and, simultaneously, to sketch the levees that might channel it. The question he poses is not whether the wave can be stopped. Suleyman is certain it cannot. The question is whether it can be navigated without drowning the institutions that hold modern civilisation together.

Section OneThe Core Argument

Suleyman's thesis rests on a deliberately paradoxical foundation. The opening chapter declares that containing the coming technological wave is not possible. The penultimate chapter insists that it must be possible. The final chapter offers ten steps that might make it so. This paradox is not rhetorical flourish but structural argument.

Technologies that are omni-use, hyper-evolving, increasingly autonomous, and capable of asymmetric impact resist every historical model of regulation that nation-states have deployed. Nuclear technology could be contained because its supply chain was narrow and expensive. AI and synthetic biology proliferate through commodity hardware and open-source code. The gap between capability and governance is widening, and Suleyman frames this gap as the defining dilemma of the twenty-first century.

Three-quarters of the text builds a rigorous case for why containment has never worked with general-purpose technologies. The remaining quarter pivots to argue that the cost of not containing these particular technologies is so severe that containment must be attempted regardless. Suleyman calls the space between these poles the "narrow path." He means it literally: the margin for error is razor-thin.

Section TwoAuthor’s Vantage

Suleyman writes from a vantage point that is both privileged and compromised. As co-founder of DeepMind, the laboratory behind AlphaFold and AlphaGo, he helped show that artificial intelligence could solve problems once considered intractable. He subsequently co-founded Inflection AI before joining Microsoft as CEO of its AI division in 2024. He holds a CBE and sits on the board of The Economist.

His co-author, Michael Bhaskar, is a writer and publisher who provides the book's narrative scaffolding. Suleyman's insider status gives the argument credibility: he has built the systems he now warns about. But it also creates a conflict of interest the reader must weigh carefully. He leads a division at Microsoft that is investing more than $140 billion in AI infrastructure in a single fiscal year. The book advocates containment while its author accelerates the very proliferation he describes.

Suleyman acknowledges this tension but does not fully resolve it. The reader should hold it in view throughout.

Insight OneTechnology as Omni-Use Force

Suleyman's most original analytical contribution is the concept of AI as an omni-use technology: one that permeates and transforms virtually every domain it touches, much as electricity remade manufacturing, communication, and urban life in the early twentieth century. This framing sets his argument apart from narrower AI-risk literature. The danger is not that a single application will cause harm. The danger is that the technology's breadth of application makes targeted regulation structurally insufficient. A tool that can diagnose disease, write legal briefs, synthesise pathogens, and coordinate autonomous weapons cannot be governed by domain-specific rules alone.

Insight TwoThe Containment Problem as Historical Pattern

Suleyman places AI within a longer narrative of technological waves: agricultural, mechanical, chemical, electrical. Each challenged existing social structures and resisted attempts at control. His key insight is that no general-purpose technology has ever been successfully contained after proliferation begins. The printing press, the internal combustion engine, and the internet all escaped the regulatory frameworks designed to manage them. He uses this historical pattern not to counsel fatalism but to establish the scale of the challenge. If containment is to work this time, it must be qualitatively different from anything attempted before.

Insight ThreeThe Four Features That Make This Wave Different

What separates this wave from its predecessors is the convergence of four features. Asymmetry: a single actor can wield disproportionate power. Hyper-evolution: iterative learning cycles accelerate development beyond human oversight. Omni-use: the technology applies across every domain. Autonomy: AI systems are beginning to conduct their own research and development. Taken together, these four features render existing governance architectures inadequate. Nuclear proliferation was containable because it lacked omni-use. Social media lacked autonomy. AI and synthetic biology combine all four for the first time in history.

Insight FourPessimism Aversion and the Failure to Prepare

Suleyman introduces the concept of pessimism aversion: the tendency of motivated reasoning to make humans systematically too optimistic about technology's trajectory. This cognitive bias, he argues, explains why societies consistently fail to prepare for technological disruption even when the evidence is visible. The term captures something that generic optimism-bias literature does not, namely a specifically technological form of denial in which the perceived benefits of a technology make its risks psychologically difficult to confront.

Suleyman frames pessimism aversion as the cultural complement to the structural containment problem. The barrier is not only institutional but perceptual.

Section SevenWhere the Argument Echoes

The book's structural claims resonate beyond their home discipline. Three parallels sharpen the argument by testing it against different terrain.

In nuclear-deterrence theory, mutual assured destruction rested on a narrow supply chain and bilateral state control. Suleyman explicitly argues that these conditions do not hold for AI. The parallel reveals something striking: his containment framework is attempting deterrence without choke-point monopoly. If nuclear containment required two superpowers to agree, AI containment requires thousands of actors across hundreds of jurisdictions. That exposes the fragility of his treaty-based proposals.

In epidemiology, zoonotic-spillover dynamics present a structurally similar problem: a pathogen jumps species barriers and proliferates before surveillance systems can respond. Suleyman's omni-use framework mirrors this spillover pattern: the technology escapes its original domain faster than governance can track it. This parallel suggests that containment strategies might benefit from public-health models of early detection and rapid response, rather than static regulatory frameworks.

In complex-systems theory, phase transitions describe moments when a system shifts abruptly from one state to another once a critical threshold is crossed. Suleyman's wave metaphor implicitly invokes these dynamics: gradual improvement in AI capability until a tipping point transforms the entire technological landscape. The parallel clarifies that the wave is not about gradual inundation. It is about sudden state-change, which implies that containment windows may be far narrower than his ten-step framework assumes.

Section EightDisciplines

If you read the synopsis once and leave with only what these four items say, you have the essential reading.

Section NineRelevance and Contextual Positioning

The book arrived in September 2023, months after ChatGPT had introduced large language models to mainstream consciousness and shortly before the first international AI Safety Summit at Bletchley Park. Since publication, Suleyman's own trajectory has illustrated the paradox he describes: he moved from advocacy to execution, joining Microsoft to build the AI systems his book urges the world to contain.

The EU AI Act, the US executive orders on AI safety, and the accelerating pace of frontier-model development have all made the book's containment question more pressing, not less. Advances in synthetic biology, the book's less-discussed second wave, continue to lower the barriers to engineering biological agents. The book reads differently in 2026 than it did in 2023. The gap between capability and governance has, if anything, widened.

Section TenWhere the Argument Strains

Suleyman's framework invites engagement from the disciplines it draws on, rather than dismissal from outside them. Four lenses bring the argument's genuine claims into sharper focus.

Political Economy

Suleyman's containment proposals rely heavily on state capacity and international cooperation, yet his own analysis demonstrates that states are losing the institutional competence required to regulate fast-moving technologies. The ten-step framework asks governments to do precisely what the book's first twelve chapters argue they cannot. This is not a contradiction Suleyman ignores; he names it. But the gap between diagnosis and prescription remains the book's most visible structural strain. The reader must decide whether the gap reflects honesty or insufficiency.

Science and Technology Studies

The wave metaphor itself invites scrutiny. Professional historians of technology, as reviewers Carl Mitcham and Martin Fuchs have noted, would criticise the metaphor as simplistic: technological change is not a natural force but a socially constructed process shaped by investment decisions, institutional incentives, and cultural narratives. By framing the wave as quasi-natural, Suleyman risks naturalising choices that are, in fact, contingent. That weakens the case for the very human agency his containment framework requires.

AI Alignment Research

Suleyman's containment framework operates at the governance and institutional level but engages only lightly with the technical alignment problem: the challenge of ensuring that increasingly capable systems pursue goals compatible with human values. His first containment step, an Apollo programme for AI safety, gestures towards this work. The book does not engage deeply with the specific technical challenges (reward hacking, mesa-optimisation, distributional shift) that alignment researchers regard as foundational. The containment architecture risks being structurally complete but technically underspecified at its most critical layer.

Environmental and Climate Scholarship

The book's treatment of AI's energy and resource footprint is notably thin. Suleyman expresses optimism that AI will help solve the climate emergency, but does not confront the tension between AI's accelerating energy consumption and planetary boundaries. Microsoft alone projects $140 billion in AI infrastructure spending. The containment framework addresses informational and biological risks in detail but largely omits the material-ecological costs of the computational substrate on which the entire wave depends.

ReferenceAdjacent Reading

ClosingClosing

The flood stories always end with a question: not whether the waters will recede, but whether anyone built an ark worth sailing. Suleyman's ten-step containment framework is his attempt at the ark. Ambitious in design, honest about the odds, structurally incomplete in ways that only become visible when examined through the lenses of the disciplines it touches.

The book's greatest contribution is not its policy proposals but its framing: the paradox that technologies which cannot be contained must be contained, and that the cost of failing to try is higher than the cost of trying imperfectly. That framing will outlast any specific regulatory proposal the book offers.

Suleyman has written a book that takes its own contradictions seriously. That is rarer than it should be in a genre saturated with either uncritical optimism or performative alarm.

If the wave is already breaking, what are you building: a levee, a surfboard, or a monument to the shore?

TakeawayWhat to Carry Away

If you read the synopsis once and leave with only what these five items say, you have the essential reading.