Daniel Levy

Daniel Levy is a French-trained AI researcher and the co-founder and president of Safe Superintelligence, the pre-product AI research company he started in June 2024 with Ilya Sutskever and Daniel Gross. Before SSI he was a member of technical staff at OpenAI from March 2022 to June 2024, where his Stanford homepage describes him as leader of the optimization team and where he is credited as optimization lead and overall vision co-lead on the GPT-4 program. As of May 2026, Levy is president of SSI following the July 3, 2025 transition in which Sutskever assumed the chief executive role and co-founder Daniel Gross departed for Meta Superintelligence Labs.

At a glance

• Education: Preparatory program at Lycée Louis-le-Grand, Paris (2010 to 2012); Diplôme d'ingénieur, École Polytechnique (2012 to 2015); MS in computer science, Stanford University (2015 to 2018), advised by Stefano Ermon; PhD in computer science, Stanford (2018 to 2021), advised by John Duchi, thesis "Advancing optimization to address the challenges of modern machine learning."
• Current role: Co-founder and president of Safe Superintelligence, since 2024 (president from July 2025).
• Key contributions: Led the optimization team at OpenAI (2022 to 2024); optimization lead and overall vision co-lead on GPT-4; first author with Duchi on "Necessary and Sufficient Geometries for Gradient Methods" (NeurIPS 2019 oral, top 36 of 6,743); equal-first author on "Large-Scale Methods for Distributionally Robust Optimization" (NeurIPS 2020); first author on "Generalizing Hamiltonian Monte Carlo with Neural Networks" (ICLR 2018).
• X / Twitter: @daniellevy__
• LinkedIn: Daniel Levy
• Personal site: ai.stanford.edu/~danilevy
• OpenReview: Daniel Asher Nathan Levy

Origins

Levy was educated in the French system. From September 2010 to June 2012 he attended the preparatory program at Lycée Louis-le-Grand in Paris on the mathematics, physics, and computer science track, the route into the entrance exams for the French Grandes Écoles. He ranked 13th nationally on the Polytechnique exam and completed the September 2012 to July 2015 Diplôme d'ingénieur at the École Polytechnique. During the program he served as a full-time teaching assistant at a Priority Action Zone school in Aulnay-sous-Bois (September 2012 to April 2013), tutoring high-school students in sciences.

In 2015 Levy moved to the United States for the Master of Science in computer science at Stanford University (September 2015 to June 2018), advised by Stefano Ermon on probabilistic models and reinforcement learning.

Career

Levy continued at Stanford for the PhD in computer science from September 2018 to December 2021, advised by John Duchi, with Percy Liang listed as second thesis advisor and Christopher Ré and Aaron Sidford on the committee. The dissertation, "Advancing optimization to address the challenges of modern machine learning," covers stochastic optimization, distributionally robust optimization, and differential privacy.

Across the Stanford period, Levy held summer research internships at Microsoft Paris (2014, Xbox Music analytics), Shift Technology in Paris (March to July 2015, bandit methods for fraud detection), and the Facebook Applied Machine Learning Group in Menlo Park (2016, bandits and reinforcement learning for text classification). He was selected for the Google Brain Residency Program in 2017 and worked with Jascha Sohl-Dickstein and Matt Hoffman on Markov-chain Monte Carlo methods, producing the ICLR 2018 paper "Generalizing Hamiltonian Monte Carlo with Neural Networks." A 2020 internship at Google Research New York with Ananda Theertha Suresh, Satyen Kale, and Mehryar Mohri produced the NeurIPS 2021 paper on user-level differential privacy. Levy was a teaching assistant for Stanford's CS229 Machine Learning in fall 2016 and EE364A Convex Optimization in winter 2021.

Levy joined OpenAI in March 2022 as a member of technical staff. His Stanford homepage lists him as leader of the optimization team. OpenAI's GPT-4 contributors page credits him as optimization lead and overall vision co-lead on the GPT-4 program, with additional roles in training-run babysitting and paper authorship. He is also listed on the GPT-4V(ision), GPT-4o, and o1 contributor pages.

On June 19, 2024, three weeks after Sutskever's May 14 announcement of his departure from OpenAI, Levy co-signed the founding statement of Safe Superintelligence with Sutskever and Daniel Gross. The announcement described SSI as "an American company with offices in Palo Alto and Tel Aviv, where we have deep roots and the ability to recruit top technical talent" and stated the mission as "one goal and one product: a safe superintelligence." Coverage in CTech reported that Levy is based in the Tel Aviv office. On the same day, Levy posted on X: "Beyond excited to be starting this company with Ilya and DG. I can't imagine working on anything else at this point in human history."

SSI raised $1 billion in September 2024 at a $5 billion valuation, with backing from NFDG, Andreessen Horowitz, Sequoia Capital, DST Global, and SV Angel. In April 2025 it raised $2 billion at a $32 billion valuation led by Greenoaks Capital, bringing cumulative funding to approximately $3 billion. On July 3, 2025, Sutskever announced that Daniel Gross had departed effective June 29 and that Sutskever had assumed the chief executive role with Levy as president; the technical team continued to report to Sutskever. Gross joined Meta Superintelligence Labs.

Affiliations

• Microsoft Paris: Intern, summer 2014 (analytics on the Cosmos big-data platform).
• Shift Technology: Intern, March 2015 to July 2015 (bandit methods for fraud detection).
• Facebook Applied Machine Learning Group: Intern, summer 2016 (bandits and reinforcement learning for text classification).
• Google Brain Residency Program: Research intern, summer 2017 (MCMC methods with Jascha Sohl-Dickstein and Matt Hoffman).
• Google Research New York: Research intern, summer 2020 (differential privacy with Ananda Theertha Suresh, Satyen Kale, and Mehryar Mohri).
• OpenAI: Member of technical staff and leader of the optimization team, 2022-03 to 2024-06.
• Safe Superintelligence: Co-founder, 2024-06-19 to present (co-founder through July 2025; president from July 2025).

Notable contributions

Levy's body of public work falls in two phases: Stanford optimization, robustness, and privacy research from 2016 through 2021, and OpenAI flagship-model training from 2022 through 2024. SSI has released no models, papers, or demonstrations as of May 2026, so Levy's record at SSI is currently the founding signature rather than research output.

• Stanford optimization research. First author with John Duchi on "Necessary and Sufficient Geometries for Gradient Methods" (NeurIPS 2019), selected for oral presentation in the top 36 of 6,743 submissions; equal-first author with Yair Carmon, Duchi, and Aaron Sidford on "Large-Scale Methods for Distributionally Robust Optimization" (NeurIPS 2020); first author on "Generalizing Hamiltonian Monte Carlo with Neural Networks" (ICLR 2018) with Matthew Hoffman and Jascha Sohl-Dickstein.
• Differential privacy. Equal-first author on "Learning with User-Level Differential Privacy" (NeurIPS 2021) with researchers from Google Research; equal-first author with Hilal Asi and Duchi on "Adapting to Function Difficulty in Private Optimization" (NeurIPS 2021).
• Distributionally robust optimization. Equal-first author on "Distributionally Robust Multilingual Machine Translation" (EMNLP 2021) with researchers at Carnegie Mellon University and Meta AI.
• OpenAI optimization-team leadership. Identified by OpenAI's GPT-4 contributions page as optimization lead, overall vision co-lead, training-run babysitter, and contributor to the GPT-4 paper. Subsequent listings on the GPT-4V, GPT-4o, and o1 contributor pages indicate continued work on flagship-model training.
• Safe Superintelligence founding (June 2024 to present). Co-founder of the pre-product research company; president since the July 2025 leadership transition. The company has secured approximately $3 billion in cumulative funding at a $32 billion valuation despite no public model, paper, or product.
• Honors. Selected for the Google Brain Residency Program in 2017 (approximately top 1 percent of applicants); oral presentation at NeurIPS 2019 (top 0.5 percent of submissions); Facebook Fellowship 2020 finalist; nominated by Stanford for the Google Fellowship.

Investments and boards

No public personal angel-investor activity on record in AI, semiconductors, datacenters, software, or energy as of May 2026. Levy's footprint in this section is concentrated in his founding and operating role at Safe Superintelligence rather than a parallel investing program.

Network

Levy's longest-running professional relationships outside OpenAI are with his Stanford optimization advisor John Duchi, his Stanford master's advisor Stefano Ermon, and the broader Stanford optimization circle of Percy Liang, Christopher Ré, Aaron Sidford, and Yair Carmon, with whom he co-authored the 2020 distributionally robust optimization paper. The Google Brain Residency cohort connects him to Jascha Sohl-Dickstein and Matthew Hoffman, his ICLR 2018 MCMC collaborators, and the 2020 Google Research privacy collaboration with Mehryar Mohri, Satyen Kale, and Ananda Theertha Suresh.

His OpenAI cohort, with whom he worked from March 2022 through June 2024, includes Ilya Sutskever, the chief scientist who became his SSI co-founder; Sam Altman, the chief executive; Greg Brockman, president and co-founder; Mira Murati, the chief technology officer through her September 2024 departure; and John Schulman, the reinforcement-learning lead through his August 2024 departure. At Safe Superintelligence, the founding cohort with Sutskever and Daniel Gross was the central operating relationship through Gross's June 2025 departure for Meta Superintelligence Labs.

Position in the field

As of May 2026, Levy occupies an unusual position among senior AI researchers. The Stanford optimization research and the NeurIPS 2019 oral presentation establish him as a credentialed optimization theorist within the John Duchi school of robust and private machine-learning optimization. The OpenAI optimization-team leadership, particularly the credited optimization-lead role on GPT-4, places him in the small group of practitioners directly responsible for training-time decisions on the largest deployed language models in 2023 and 2024.

The SSI co-founder and president role situates him alongside Sutskever as the public face of the most distinctive funding-and-strategy posture among AI labs: a $32 billion valuation, no public product, no published research, no website beyond a single page of plain text, and a stated commitment to release nothing until the company achieves safe superintelligence. The thesis underwriting the $3 billion in funding rests on Sutskever's research credentials and on the credentials of co-founders including Levy.

Levy's public profile is concentrated on the Stanford homepage, the @daniellevy__ account on X, and a small number of conference talks tied to his Stanford research period. His role within SSI has not been publicly described in technical specifics, consistent with the company's no-disclosure posture.

Outlook

Open questions over the next 6 to 18 months:

• First Safe Superintelligence release. Whether SSI produces any public artifact, paper, or demonstration, and Levy's specific role on a first release.
• Public technical role at SSI. Whether Levy's specific research direction beyond the president title is publicly described in 2026 or 2027.
• Next funding round. Whether SSI raises at a higher valuation or holds at $32 billion as the no-product runway extends.
• Tel Aviv research presence. Whether the SSI Tel Aviv office grows materially in 2026 and whether Levy remains based there.
• Public commentary cadence. Whether Levy increases his English-language conference, talk, or podcast presence beyond the founding announcement and the @daniellevy__ X account.

Sources

• Daniel Levy. Stanford AI Lab homepage with full education history, advisors, internships, publications, and the listing of his role as leader of the optimization team at OpenAI.
• Daniel Levy CV. Stanford-hosted curriculum vitae covering education from 2010 to 2021, professional experience through OpenAI from March 2022, the full publication record, and the honors and invited-talks list.
• Daniel Asher Nathan Levy. OpenReview profile with publication record, affiliations, and PhD-advisor relationship to John Duchi.
• Advancing optimization for modern machine learning. Stanford library catalog entry for Levy's December 2021 PhD dissertation, with John Duchi and Percy Liang listed as advisors and Christopher Ré and Aaron Sidford on the committee.
• Safe Superintelligence Inc.. The single-page company website with the founding mission statement, signed by Sutskever, Gross, and Levy.
• SSI updates page. Company updates including the September 2024 funding announcement and the July 2025 leadership transition message naming Levy as president.
• OpenAI co-founder Ilya Sutskever founding new AI company with offices in Tel Aviv. CTech coverage of the SSI launch, identifying Levy's France-Stanford-OpenAI background and his Tel Aviv-based location at SSI.
• Safe Superintelligence Inc.. Wikipedia entry on the company with co-founder identification, funding history, and the July 2025 leadership transition.
• Necessary and Sufficient Geometries for Gradient Methods. The 2019 NeurIPS oral paper by Levy and Duchi.
• Large-Scale Methods for Distributionally Robust Optimization. The 2020 NeurIPS paper by Levy, Carmon, Duchi, and Sidford.
• Generalizing Hamiltonian Monte Carlo with Neural Networks. The 2018 ICLR paper by Levy, Hoffman, and Sohl-Dickstein.
• Learning with User-Level Differential Privacy. The 2021 NeurIPS paper on user-level privacy from the Google Research collaboration.
• GPT-4 contributions. OpenAI's contributors page identifying Levy as optimization lead, overall vision co-lead, training-run babysitter, and paper contributor on GPT-4.
• GPT-4V(ision) contributions, GPT-4o contributions, and OpenAI o1 contributions. Subsequent OpenAI contributor pages listing Levy.
• Daniel Levy on LinkedIn and @daniellevy__ on X. Personal social profiles.
• Feature image: text-mode card generated via scripts/make_lab_card.py; no credit-cleared portrait was located in May 2026.