Safe Superintelligence Inc.: what is it trying to build?

This page is part of our series on who is building superintelligence. It is short because the public record is short.

What SSI says

SSI’s website, checked on 3 October 2026, consists of a single statement. Its key sentences are:

• “Superintelligence is within reach.”
• “Building safe superintelligence (SSI) is the most important technical problem of our time.”
• It describes itself as “the world’s first straight-shot SSI lab, with one goal and one product: a safe superintelligence”.
• “We approach safety and capabilities in tandem, as technical problems to be solved through revolutionary engineering and scientific breakthroughs.”
• It plans “to advance capabilities as fast as possible while making sure our safety always remains ahead”, with safety, security and progress “insulated from short-term commercial pressures”.

It describes itself as an American company with offices in Palo Alto and Tel Aviv.

“Straight-shot” is the distinctive idea. Other frontier developers release a series of products, earning revenue and learning from each. SSI says instead that safe superintelligence is “our mission, our name, and our entire product roadmap”. Sutskever has since softened this slightly: in a November 2025 interview he said there was “merit” to the straight-shot approach, but that “even in the straight shot scenario, you would still do a gradual release”.

What it means by safe superintelligence

SSI has not published a definition of either word. Its website does not say what capability would count as superintelligence, or what “safe” would require. The nearest thing to an explanation is Sutskever’s own interviews. In the November 2025 conversation he described the goal as “AI that’s robustly aligned to care about sentient life specifically”, and said SSI had “a different technical approach” that he considered worth pursuing, without describing it.

Because the company says “superintelligence” without qualification, and its founder is a researcher steeped in the field’s debates, the word is probably meant in something close to the research sense this site uses: a system that greatly exceeds the best humans in virtually all domains (explained here). That is an inference, not something SSI has stated. It is quite different from Meta’s “personal superintelligence” (explained here).

What its founder has said about the technology

Sutskever has given a handful of public talks and interviews since founding SSI. The views relevant to its goal:

• At the NeurIPS conference in December 2024 he said that pre-training “as we know it will unquestionably end”, because the supply of training data is limited: “there’s only one internet”.
• In November 2025 he said the period from 2020 to 2025 had been “the age of scaling” and that the field was going “back to the age of research again, just with big computers”. He said current models “generalize dramatically worse than people”.
• Asked for his forecast for a system that “can learn as well as a human and subsequently, as a result, become superhuman”, he said “5 to 20” years.

These are the views of SSI’s chief executive. They indicate the company is betting on new research ideas rather than simply scaling existing methods, but they say nothing verifiable about what SSI has achieved. Forecasts from other developers are on our timelines page.

People, funding and compute

SSI’s own updates page records three events. In September 2024 it announced it had raised $1 billion from investors including Andreessen Horowitz and Sequoia. On 3 July 2025 Sutskever wrote that Gross had left on 29 June, that he himself was now chief executive and Levy president, and that “we have the compute, we have the team, and we know what to do”. Gross joined Meta, according to press reports. On 26 July 2026 (27 July in Nvidia’s announcement) SSI announced a partnership with Nvidia that it said would let it “scale our compute by 10x”.

Beyond that, the record comes from partners and the press. Google Cloud said in April 2025 that SSI was using its TPU chips for research. Press reports in 2025 described a further $2 billion round at a valuation of about $32 billion, and in November 2025 Sutskever said SSI had raised $3 billion in total. Neither company disclosed the size of Nvidia’s 2026 investment; Bloomberg reported it at up to $5 billion.

What is not known

As of 3 October 2026 SSI had released no model, product, paper or benchmark result, and this site found no independent evaluation of its work. Its size, its technical approach, how it intends to make a superintelligence safe, and how it would decide that one had been built are not public. It did not sign the White House Accord on Super Intelligence of 29 September 2026, which was signed by six other companies’ leaders.

Social-media posts in early October 2026 hinted at an imminent announcement; this site has not been able to connect them to the company. This page will be updated if SSI publishes anything.

The questions SSI says it is working on, making a far more capable system that stays aligned and controllable, are set out on our pages on alignment and control.