In May 2024, Ilya Sutskever — OpenAI co-founder and one of the intellectual architects of the models that made ChatGPT possible — left the company following the turbulent November 2023 and months of public silence. In June 2024, he founded Safe Superintelligence Inc. (SSI) alongside Daniel Gross and Daniel Levi. The proposal was radical in its simplicity: build only one thing, safe superintelligence, with no product, no revenue, and no distraction.
In 2026, the first signs of what SSI is developing are starting to circulate — and they are proving interesting enough to force recalibrations at the major labs.
What We Know About SSI
SSI raised US$ 1 billion in June 2024, led by Andreessen Horowitz, Sequoia, DST Global, and SV Angel, at a US$ 5 billion valuation — with no product, no revenue, and solely the thesis that Sutskever and his team would know how to build superintelligence safely. By 2026, the company has not released any public products, has not published papers, and has made no substantive statements about technical progress.
What circulates is inference: according to analyst predictions published by Forbes in December 2025, details of SSI's approach would begin to leak to the public in 2026 — "a research agenda novel and promising enough to prompt the major labs to recalibrate their own roadmaps."
Sutskever and What He Believes
To understand what SSI is likely building, one must understand what Sutskever believes about AI. In his last major public appearance before leaving OpenAI, he stated at NeurIPS 2023: "Neural networks are slightly conscious." In communications from his time at OpenAI, he expressed deep concern about alignment — the idea that more capable models might have goals misaligned with humans.
The most likely candidates for SSI's research agenda: recursive self-improvement (AI systems capable of improving themselves, which then improve even more capable systems) or continuous learning (systems that learn continuously during interaction with the world, unlike current models that remain static post-training). Both address structural limitations of current LLMs and are frontier research areas in 2026.
Why the Field Is Paying Attention
SSI's importance lies not in what it has produced — it is about who is there. Sutskever is a co-author of the GPT-2 paper, GPT-3, InstructGPT, and the original RLHF paper that laid the groundwork for modern alignment. His departure from OpenAI was partially motivated by disagreements over deployment speed versus safety — disagreements that became public in the November 2023 episode when OpenAI's board attempted to fire Sam Altman.
SSI's implicit thesis: the labs leading the field are moving too fast without solving the alignment problem first. And the only way to build safe superintelligence is through an organization that lacks short-term commercial incentives pushing in the opposite direction.
The Business Model Risk
SSI's model — no product, no revenue, pure research — is financially sustainable only with venture capital and for as long as investors believe the thesis holds value. With an initial US$ 1 billion at a burn rate typical for an elite research lab, SSI has a runway of a few years.
The question the field is watching: if the details leaking out in 2026 confirm a research agenda genuinely different from the paths OpenAI, Anthropic, and DeepMind are pursuing — and promising enough to justify the attention — SSI could become the most influential lab that never released a public product. If not, it will be remembered as one of venture capital's most expensive bets on AI research.
