# Ilya Sutskever Ilya Sutskever (b. 1986) is a **Russian-Israeli-Canadian-American computer scientist** whose career has placed him at the center of the deep learning revolution (2012-2016), the founding of OpenAI (2015-2024), and — since June 2024 — his new company **Safe Superintelligence Inc. (SSI)**. A student of Geoffrey Hinton at the University of Toronto, Sutskever was a co-author on the landmark 2012 **AlexNet paper** that inaugurated the deep-learning revolution in computer vision, subsequently a research scientist at Google Brain (2012-2015) where he co-authored foundational papers on sequence-to-sequence learning that became the basis for modern neural machine translation, and then a **co-founder and Chief Scientist of OpenAI** (2015-2024). Sutskever was one of the four directors of OpenAI's nonprofit board who **voted to remove Sam Altman as CEO on November 17, 2023** — the vote that triggered the five-day governance crisis — and subsequently publicly regretted his role in the vote and departed OpenAI in May 2024. In June 2024, he announced **Safe Superintelligence Inc. (SSI)**, a new AI company with the explicit mission of building safe superintelligence as its sole product. Sutskever matters for the KB for three reasons. First, his **technical contributions are foundational** to the modern AI era — AlexNet (2012), sequence-to-sequence learning (2014), contributions to AlphaGo (2016), the Transformer-era scaling that OpenAI pursued (2019-2024) all bear his research imprint. Among the senior OpenAI leadership, Sutskever was the most substantively technical figure — the scientific director who made many of the key research calls during OpenAI's GPT-series development. Second, his **November 2023 vote to fire Altman** is structurally central to the KB's OpenAI-crisis documentation — Sutskever's vote, his subsequent reversal (signing the November 20 employee letter calling for Altman's reinstatement), and his eventual May 2024 departure represent one of the most documented trajectories of senior AI-safety-motivated resistance to commercially-oriented AI development, with eventual accommodation producing resignation rather than sustained opposition. Third, his **June 2024 founding of Safe Superintelligence Inc.** is one of the most significant post-OpenAI institutional moves — SSI's explicit commitment to building safe superintelligence as its only product, its initial $1 billion funding round, and its $5 billion valuation at founding, make it one of the principal institutional tests of whether AI-safety-focused development can maintain its commitment to safety under competitive capital-market pressures. ## Background - Born 1986 in Nizhny Novgorod, Soviet Union (now Russia) - Family immigrated to Jerusalem, Israel, at age 5; Sutskever grew up in Israel - Family immigrated to Toronto, Canada, in 2002 when Sutskever was a teenager - BS, MS, and PhD from the University of Toronto (2005-2013; advisor Geoffrey Hinton) - Co-founder of DNNresearch (Hinton's startup that was acquired by Google in March 2013) - Research scientist, Google Brain (2013-2015) - Co-founder and Chief Scientist, OpenAI (December 2015 – May 2024) - Co-lead, OpenAI Superalignment Team (July 2023 – May 2024) - Founder, Safe Superintelligence Inc. (June 2024 – present) ## Pre-OpenAI Technical Work (2005-2015) Sutskever's pre-OpenAI research career included several landmark contributions: ### AlexNet (2012) The 2012 paper *ImageNet Classification with Deep Convolutional Neural Networks* — co-authored with Alex Krizhevsky (first author) and Geoffrey Hinton — is widely considered the landmark paper that inaugurated the modern deep learning revolution. The paper demonstrated that a deep convolutional neural network could achieve dramatic improvements over prior approaches on the ImageNet benchmark. Its success led directly to: - The widespread adoption of deep learning in computer vision - The recognition that GPU-accelerated training of deep networks was the principal path forward for the field - The subsequent hiring of Krizhevsky, Sutskever, and Hinton by Google in March 2013 through DNNresearch ### Sequence-to-Sequence Learning (2014) The 2014 paper *Sequence to Sequence Learning with Neural Networks* (Sutskever, Vinyals, Le) established the sequence-to-sequence framework that became the basis for modern neural machine translation, ChatGPT's core architecture, and many other generative modeling systems. The paper's core insight — that sequential processing problems could be framed as mapping one sequence to another through an encoder-decoder architecture — was foundational for the subsequent Transformer-era work. ### AlphaGo contributions Sutskever made early contributions to what became DeepMind's AlphaGo work during his Google Brain tenure. ## OpenAI Co-Founding and Chief Scientist Role (2015-2024) Sutskever was one of the core co-founders when OpenAI was launched on December 11, 2015. He was the **Chief Scientist** from founding through his May 2024 departure. His principal contributions: ### Scaling strategy Sutskever was the principal technical advocate within OpenAI for the "scaling hypothesis" — the claim that substantial AI capability gains could be achieved primarily through scaling existing neural network architectures with more compute, more data, and more parameters. This strategic bet, which was the research direction of GPT-1 (2018), GPT-2 (2019), GPT-3 (2020), and the subsequent GPT-4 and beyond, was substantially driven by Sutskever's conviction that scaling would produce meaningful capability emergence. ### GPT research program Sutskever was the senior technical figure overseeing the GPT research program across OpenAI's existence. The progression from GPT-1's demonstration of the approach through GPT-4's broad capability across reasoning tasks was the principal research achievement of OpenAI's first decade. ### Superalignment team (July 2023) In July 2023, Sutskever and Jan Leike were named co-leads of OpenAI's new **Superalignment team**, a research program committed to solving the technical problem of aligning superintelligent AI systems within four years. The team was committed 20% of OpenAI's compute resources and was intended to be the organization's principal AI safety research effort. ## The November 2023 Governance Crisis Sutskever's role in the November 2023 OpenAI governance crisis is one of the most documented and most discussed elements of the crisis: ### The Firing Vote (November 17, 2023) Sutskever, along with board members Adam D'Angelo, Helen Toner, and Tasha McCauley, voted to fire Sam Altman as CEO on Friday afternoon November 17. The board's stated reasoning — that Altman was "not consistently candid" with the board — has been elaborated in subsequent reporting as including concerns about Altman's handling of AI safety research and his efforts to reduce Toner's board influence. ### The Reversal (November 20, 2023) On Sunday November 19 — approximately 48 hours after voting to fire Altman — Sutskever publicly posted on Twitter/X: "I deeply regret my participation in the board's actions. I never intended to harm OpenAI. I love everything we've built together and I will do everything I can to reunite the company." Sutskever then signed the November 20 employee letter calling for Altman's reinstatement. ### The Aftermath Sutskever did not return to the OpenAI board under the restructured governance. Reports indicated Sutskever remained at OpenAI with a reduced public profile through early 2024, principally working on the Superalignment team. ## Departure (May 2024) and Superalignment Dissolution On May 14, 2024, Sutskever announced his departure from OpenAI. His departure was approximately concurrent with Jan Leike's (May 15), and the Superalignment team was effectively dissolved within weeks. The stated reasons in subsequent reporting included: - Concerns that OpenAI was not delivering on the 20%-of-compute commitment for Superalignment research - Broader concerns about the tension between OpenAI's commercial trajectory and its safety mission - Personal considerations about building a new organization with clearer safety focus Jan Leike's subsequent public statement (May 17, 2024) explicitly criticized OpenAI's direction, stating "safety culture and processes have taken a back seat to shiny products." Sutskever's public statements were more measured but were understood as consistent with Leike's framing. ## Safe Superintelligence Inc. (June 2024) On June 19, 2024, Sutskever announced **Safe Superintelligence Inc. (SSI)** with co-founders [[Daniel Gross]] (former Apple AI lead) and [[Daniel Levy]] (former OpenAI researcher). The company's announcement emphasized: - **Single-product focus**: Building safe superintelligence as the only product - **No commercial distractions**: Explicit commitment to avoiding near-term revenue generation - **Dual offices**: Palo Alto and Tel Aviv - **Safety-first capability development**: Research approach that treats safety and capability as co-developed rather than as later-stage safety retrofitting SSI's funding trajectory: - **September 2024**: $1 billion seed at ~$5 billion valuation ([[Sequoia Capital]], [[Andreessen Horowitz]], [[DST Global]], SV Angel, [[NFDG]]) - **Spring 2025**: ~$2 billion led by [[Greenoaks]] at $30–32 billion (reporting varies on both the month and the figure) - **July 27, 2026**: $5 billion from [[NVIDIA]] plus Vera Rubin platform access — see [[NVIDIA-Safe Superintelligence Investment]] **Leadership change.** [[Meta]] attempted to acquire SSI outright in the first half of 2025; Sutskever declined. [[Mark Zuckerberg]] then hired Gross and [[Nat Friedman]] into [[Meta Superintelligence Labs]] instead. Sutskever became CEO on July 3, 2025, with [[Daniel Levy]] as president. **The research thesis.** In November 2025 Sutskever argued publicly that the industry had passed from the "age of scaling" into an "age of research" — that pre-training as practiced since GPT-3 had exhausted its easy returns and that further gains would come from ideas rather than compute. This is a notable reversal for the person most identified with the [[Scaling Hypothesis]], and it is the intellectual justification for SSI's small-team, no-product structure. The SSI founding remains one of the most consequential post-OpenAI moves of the period. Whether SSI can maintain safety focus at the capital-market scale it has attained is a live question — one complicated by the fact that its two largest strategic backers, [[NVIDIA]] and [[Alphabet]], are the companies with the most direct commercial interest in unconstrained AI compute. Full treatment: [[Safe Superintelligence]]. ## Connection to the KB Ecosystem - **[[OpenAI]]**: Co-founder and Chief Scientist; principal technical figure through May 2024. - **[[Sam Altman]]**: Co-founder partnership; voted to fire him in November 2023, subsequently reversed. - **[[Greg Brockman]]**: Co-founder partnership; parallel senior role. - **[[Elon Musk]]**: OpenAI co-founder; early supporter. - **[[Geoffrey Hinton]]**: PhD advisor; co-author on AlexNet. - **[[AI Safety]]**: Principal senior technical figure advocating for AI safety prioritization within OpenAI; continued advocate through SSI. - **November 2023 OpenAI governance crisis**: Principal figure as board member who voted for Altman removal, then reversed. - **Deep learning revolution**: AlexNet, seq2seq contributions are foundational. - **[[Safe Superintelligence]]**: Founder and, since July 2025, CEO; one of the most consequential post-OpenAI institutional moves. ## See Also - [[Safe Superintelligence]] - [[Daniel Gross]] - [[Daniel Levy]] - [[Sam Altman]] - [[Greg Brockman]] - [[Elon Musk]] - [[Geoffrey Hinton]] - [[OpenAI]] - [[AI Safety]]