Bio

Laurent Sifre is a French computer scientist, co-founder and chief technology officer of H, the Paris-headquartered agentic-AI company he established in 2023 with Charles Kantor, Karl Tuyls, Daan Wierstra, and Julien Perolat. He is a co-author of the 2016 Nature paper Mastering the game of Go with deep neural networks and tree search, the principal AlphaGo publication, and the 2017 Nature paper Mastering the game of Go without human knowledge on AlphaGo Zero. As of May 2026, he leads the technical and research direction at H following the June 2025 chief-executive transition to Gautier Cloix and the public-beta launch of the Runner H, Surfer H, and Tester H product suite alongside the Holo-1 open-weights model.

At a glance

• Education: Engineer's degree, École Polytechnique; Master's in mathematics, vision, and learning at École Normale Supérieure Paris-Saclay; PhD in applied mathematics, École Polytechnique (2014), advised by Stéphane Mallat on rigid-motion scattering for image classification.
• Current role: Co-founder and Chief Technology Officer of H since 2023 (founding), CTO from June 2025.
• Past roles: Principal Research Scientist at Google DeepMind, August 2014 to early 2024; research intern at Google Brain, 2013.
• Key contributions: co-author of the 2016 AlphaGo and 2017 AlphaGo Zero Nature papers; design and implementation of the AlphaGo Zero reinforcement-learning algorithm; contributions to AlphaFold, AlphaStar, MuZero, and Google DeepMind's large-language-model program; co-founder of H.
• X / Twitter: @laurentsifre
• LinkedIn: Laurent Sifre

Origins

Sifre completed an Engineer's degree at École Polytechnique in Paris and a master's in mathematics, vision, and learning at École Normale Supérieure Paris-Saclay before entering the doctoral program at École Polytechnique under Stéphane Mallat, the founder of the wavelet-and-scattering school of applied mathematics. The PhD, completed in 2014, addressed rigid-motion scattering for image classification, a research direction within the broader Mallat-school program of building invariant image representations through scattering transforms. The thesis applied the framework to texture and image classification benchmarks.

In 2013, between thesis years, he completed a research internship at Google Brain in Mountain View. The internship placed him inside the Google research environment during the early deep-learning consolidation period and led to the August 2014 hire at DeepMind.

Career

Sifre joined DeepMind in August 2014 as a research scientist, with the title progression to Principal Research Scientist over a roughly nine-and-a-half-year tenure. The principal research artifacts from the DeepMind period span the AlphaGo, AlphaGo Zero, AlphaZero, AlphaFold, AlphaStar, and MuZero programs, with subsequent contributions to the large-language-model line that became Gemini.

The 2016 Nature paper Mastering the game of Go with deep neural networks and tree search introduced AlphaGo, the first program to defeat a top professional Go player without handicap. The paper, with David Silver, Aja Huang, Chris Maddison, Arthur Guez, Sifre, and others as co-authors, set the program against Fan Hui in 2015 and described the deep-neural-network plus Monte Carlo tree search architecture that defeated Lee Sedol 4 to 1 in March 2016. The 2017 Nature paper Mastering the game of Go without human knowledge introduced AlphaGo Zero, which mastered Go from self-play with no human game data. Sifre was a co-author on both papers and, per public statements at H Company, designed and implemented the reinforcement-learning algorithm in AlphaGo Zero.

Subsequent contributions in the DeepMind period included the AlphaZero generalization to chess and shogi (2018), the AlphaStar real-time-strategy program in StarCraft II, the MuZero model-based extension that learned without environment rules, and the AlphaFold protein-structure-prediction program. He was an author on the Chinchilla scaling-law paper (March 2022), the principal Google DeepMind research artifact on compute-optimal large-language-model training, alongside Arthur Mensch and a 22-author team led by Jordan Hoffmann and Sebastian Borgeaud.

In early 2024 Sifre left Google DeepMind to co-found H in Paris with Charles Kantor, Karl Tuyls, Daan Wierstra, and Julien Perolat. The founding thesis was the pursuit of "frontier action models," foundation models trained for autonomous task execution rather than conversation. The May 2024 $220 million seed round, led by Accel with participation from Amazon, UiPath, Bpifrance, Innovation Endeavors, FirstMark, and Bernard Arnault, set an early benchmark for European agentic-AI fundraising.

The first year was disrupted by the late-2024 departures of three of the five co-founders (Tuyls, Wierstra, and Perolat) over what the company described as "operational differences." Sifre and Kantor remained on the founding team. Sifre transitioned into the Chief Technology Officer role and continued to lead the technical direction through the November 2024 Runner H closed-beta launch, the June 2025 public release of the Runner H, Surfer H, and Tester H product suite, the launch of the Holo-1 open-weights visual-language model under Apache 2.0, and the June 2025 chief-executive transition from Kantor to Gautier Cloix, formerly the managing director of Palantir's France unit.

Affiliations

• École Polytechnique: Doctoral candidate in applied mathematics, advised by Stéphane Mallat; PhD awarded 2014.
• Google Brain: Research Intern, 2013.
• Google DeepMind: Research Scientist and Principal Research Scientist, August 2014 to early 2024.
• H: Co-founder, 2023 to present; Chief Technology Officer, June 2025 to present.

Notable contributions

• Mastering the game of Go with deep neural networks and tree search (Nature, January 2016). Co-author of the original AlphaGo paper. AlphaGo defeated 9-dan professional Lee Sedol 4 to 1 in March 2016 in the Seoul match watched by an estimated 200 million viewers, the first program to beat a top professional Go player without handicap.
• Mastering the game of Go without human knowledge (Nature, October 2017). Co-author of the AlphaGo Zero paper. Per his public framing at H Company, he designed and implemented the reinforcement-learning algorithm in AlphaGo Zero.
• AlphaZero, AlphaStar, MuZero, AlphaFold contributions. Subsequent contributions across the DeepMind research lines through the late 2010s and early 2020s.
• Training Compute-Optimal Large Language Models (March 2022). Co-author on the Google DeepMind Chinchilla paper that revised compute-optimal scaling for large language models, alongside Mensch and a 22-author team.
• H founding (2023). Co-founded the Paris-headquartered agentic-AI company.
• Runner H, Surfer H, Tester H product suite, and Holo-1 open-weights model (June 2025). Led the technical direction through the public release of the agentic-AI product suite and the open-weights visual-language model on Hugging Face under Apache 2.0.

Position in the field

As of May 2026, Sifre occupies a distinctive position as the senior research-and-engineering leader at H. His co-authorship on the AlphaGo and AlphaGo Zero papers places him inside the technical lineage of contemporary deep reinforcement learning, alongside David Silver, Aja Huang, and the broader DeepMind-AlphaGo research cohort.

The continuation of the H Chief Technology Officer role through the post-departure period and the chief-executive transition reframed the company's narrative around Sifre as the principal research-and-engineering anchor. Industry coverage has characterized the H trajectory through 2025 and into 2026 as one in which Sifre's continued technical leadership has been a stabilizing factor for the company through structural turbulence. The competitive question for H is whether the agentic-AI capability claims and the Holo-1 open-weights line translate into commercial customer traction at the implied seed valuation.

Outlook

Open questions over the next 6 to 18 months:

• Successor agentic models. Continued capability disclosures from the H product suite and any flagship-model successors beyond Runner H, Surfer H, and Tester H.
• Holo-1 successor releases. Continued open-weights releases on the visual-language line under Apache 2.0.
• Series A or growth-round close. H's next institutional fundraising and the implied valuation given the leadership churn and product execution since seed.
• Senior research recruiting. Continued ex-DeepMind, ex-Meta, and ex-Mistral hires under Sifre's technical direction.
• Cloix-Sifre operating partnership. The durability of the chief-executive-and-CTO operating partnership through 2026.

Sources

• Mastering the game of Go with deep neural networks and tree search. The 2016 Nature paper introducing AlphaGo, with Sifre as a co-author.
• Mastering the game of Go without human knowledge. The 2017 Nature paper on AlphaGo Zero, with Sifre as a co-author.
• Laurent Sifre at Slush. Slush profile detailing the H Company role.
• From DeepMind to the Frontier of Machine Autonomy. VAST Data profile covering Sifre's DeepMind tenure and H Company founding.
• Laurent Sifre on Chess Programming wiki. Reference page on the AlphaGo and AlphaGo Zero co-authorship.
• Laurent Sifre on DeepAI. DeepAI biography aggregating publication history.
• Training Compute-Optimal Large Language Models. The March 2022 Google DeepMind Chinchilla paper with Sifre as a co-author.
• TechCrunch: French AI startup H raises $220M seed round. Seed-round announcement.
• BusinessWire: H Company Launches Next-Generation Autonomous AI Agents. June 2025 product-suite launch announcement.