The Coming Wave — Suleyman and Bhaskar (2023)

The Coming Wave is a 2023 book by Mustafa Suleyman, co-founder of DeepMind and Inflection (and CEO of Microsoft AI as of 2024), written with Michael Bhaskar and published by Crown in September 2023. Its central claim is that a convergent "wave" of AI and synthetic biology will reshape power more radically than any prior technology, that this wave cannot be paused, and that the defining policy problem is what Suleyman terms containment — managed development that lets benefits flow broadly while preventing catastrophic harms.

Authors: Mustafa Suleyman (co-founder of DeepMind, Inflection; CEO of Microsoft AI as of 2024) with Michael Bhaskar Published: September 2023 (Crown)

Core argument

Suleyman and Bhaskar argue that a technological wave combining AI and synthetic biology is arriving that will reshape power more radically than any prior wave. They frame the defining question as containment: ensuring that benefits flow broadly while catastrophic harms are prevented. Previous waves — agriculture, industrial, nuclear, electric, digital — were, in their account, all eventually contained through nation-state capacity. The authors argue that this wave is structurally harder to contain because of four properties:

1. Omni-use — a single capability (protein design, autonomous code generation, persuasion) has civilian, military, commercial, and weapons-of-mass-destruction applications.
2. Hyper-evolution — iteration cycles measured in months rather than the decades of prior waves.
3. Asymmetric impact — a small group of actors, or a single model, can produce globally-scaled harm.
4. Increasing autonomy — systems act in the world with diminishing human oversight.

In the authors' view, the combination of these properties has no precedent, and prior containment frameworks — nuclear non-proliferation, export controls, pharmaceutical regulation — are each insufficient individually; the wave, they argue, requires a synthesis of all of them.

The "pessimism aversion trap"

Suleyman argues that those closest to the technology, including builders and investors, systematically under-rate catastrophic scenarios because pessimism is culturally and professionally penalized within the technology sector — a pattern he calls the "pessimism aversion trap." He frames the book as a deliberate act of pessimism from an insider: a frontier-lab co-founder arguing for heavier containment on the grounds that he understands how fast capabilities are moving.

Containment strategy — 10-point framework

Suleyman proposes a containment strategy organized around ten elements:

1. Safety — technical and operational research (interpretability, red-teaming).
2. Audits — third-party technical and behavioral audits.
3. Choke points — targeting compute, data, and talent as controllable leverage.
4. Makers — holding lab leadership personally responsible.
5. Businesses — market-based incentives that reward safety.
6. Governments — capacity-building inside regulators.
7. Alliances — international coordination (G7, AISI-equivalents).
8. Culture — norms within the technology community that permit pessimism.
9. Movements — public pressure that makes containment politically profitable.
10. Coherence — synthesis across the above, which the authors describe as the hardest and most important element.

Relation to later frameworks and developments

The book's containment framework anticipates several later AI-governance instruments. Its audits, choke-points, and makers items closely echo the "frontier framework" structure of Anthropic's Responsible Scaling Policy (Version 3.1) and OpenAI Preparedness Framework V.2. Its international-coordination element parallels the The Bletchley Declaration (AI Safety Summit, 1–2 November 2023), the Frontier AI Safety Commitments (Seoul, 2024), and the Paris AI Action Summit Declaration (2025). The choke-point strategy applied to compute corresponds to the BIS Framework for AI Diffusion — Interim Final Rule (Jan 13, 2025) and the export-control analysis in CSIS — DeepSeek, Huawei, Export Controls, and the Future of the U.S.-China AI Race (Allen, March 2025). The regulatory-capacity-building element corresponds to AISI work described in UK AISI Frontier AI Trends Report 2025, and the pessimism-aversion critique shares lineage with FLI — Pause Giant AI Experiments: An Open Letter.

Several of the book's claims have been borne out by subsequent events. The omni-use characterization is reflected in agent-enabled cyberattacks (Source: Raw Sources/Mandia - AI Agent Cyberattacks Warning 2025.md) and in the cybersecurity capabilities attributed to Anthropic's Claude Mythos model (Source: nytimes.com). The choke-point strategy targeting compute and talent corresponds to the export-control and industrial-policy approaches pursued by the Biden and Trump administrations. The internal-lab pessimism-aversion thesis is consistent with METR Review of Anthropic's Sabotage Risk Report (Opus 4.6), which found external reviewers often more cautious than the labs themselves.

Other claims have aged less well or remain contested. The AI–synthetic-biology convergence has become less symmetric, as the AI side has advanced faster than synthetic biology. Containment feasibility has been challenged by open-weight model proliferation (DeepSeek, Qwen, Kimi K2), which has further weakened containment leverage since 2023. Suleyman's own trajectory is in tension with the book's containment thesis: he wrote it while leading Inflection and has since become CEO of Microsoft AI.

Relationships

• supports: Statement on AI Risk (CAIS) / Superintelligence: Paths, Dangers, Strategies — Bostrom (2014) — same overall concern, framed in more policy-operational and less metaphysical terms.
• supports: Anthropic's Responsible Scaling Policy (Version 3.1) / OpenAI Preparedness Framework V.2 — frontier frameworks implementing Suleyman-style containment.
• related: Introducing LawZero (Bengio) — Bengio's initiative shares the wave/containment framing.
• contradicts: AI as Normal Technology / AI Snake Oil — Narayanan and Kapoor (2024) / The Digitalist Papers (Stanford, Volumes 1–2) (Cochrane essay) — reject the regulatory-imperative framing.
• related: The Digitalist Papers (Stanford, Volumes 1–2) — referenced by several Vol. 2 essays.