Superintelligence: what it is, when it will arrive and what it changes for those who lead a company

Artificial superintelligence is a hypothetical system capable of surpassing human cognitive abilities across a broad range of domains. It does not exist today: although the models we use can achieve extraordinary results in specific tasks, they cannot be described as “superintelligent”, only as general.

On 22 September 2026, before the UN General Assembly, Donald Trump announced that US government documents will use “super intelligence” instead of “artificial intelligence”. He gave today’s AI a new name, using an expression that in research denotes a very different, still hypothetical technology.

His reason for doing so is political, and we have analysed it here. In this article, instead, we look at the real, technical distinction: what a superintelligence would actually be, when it might arrive and which decisions already concern those who govern AI within a company.

What artificial superintelligence is

The term artificial superintelligence (ASI for short) refers to an artificial intelligence system that surpasses human cognitive capabilities in every domain. According to the philosopher Nick Bostrom, who formalised the concept in his 2014 book Superintelligence, an ASI is

“An intellect that is much smarter than the best human brains in practically every field, including scientific creativity, general wisdom and social skills”.

This definition is philosophical, economic and strategic all at once. Because a system with those characteristics would be far more than a tool. It would be an autonomous agent capable of solving problems no human being can tackle alone, of accelerating scientific research, of rewriting entire business models.

The three levels of artificial intelligence: from Narrow AI to superintelligence

Understanding superintelligence requires placing it within the evolutionary sequence of AI systems. The technical literature distinguishes three levels.

The systems available today can achieve highly advanced results in specific tasks, but there is no consensus that they have reached a general intelligence comparable to human intelligence. AGI describes this hypothetical general capability; superintelligence marks a further threshold, beyond human capabilities.

Forecasts on timing diverge and also depend on the criteria used to recognise each threshold. A date announced by a lab is a forecast, not a milestone already reached (IBM).

Who, what, where, when, why: the 5 Ws of superintelligence

Who is developing it

OpenAI, Anthropic, DeepMind, xAI, Meta AI. The world’s leading laboratories are investing tens of billions of dollars in pursuit of AGI as an intermediate step towards superintelligence. The Stargate project, announced on 21 January 2025, is a joint venture between OpenAI, SoftBank and Oracle with the explicit aim of building the computing infrastructure needed for next-generation AI systems. Meta has set up a dedicated team, TBD Lab, funded with tens of billions of dollars, with the goal of achieving superintelligence.

What it is (and what it is not)

Superintelligence refers to a hypothetical system with cognitive capabilities far exceeding those of humans. Recursive self-improvement is one of the possible trajectories discussed by researchers; it is neither a feature already observed in current systems nor a guaranteed consequence of the definition. IBM defines artificial superintelligence as a system capable of cross-domain reasoning, continuous learning and independent decision-making on any problem a human being might face, and of solving it faster and more accurately.

Where it will arrive first

Scientific research, computational biology, cybersecurity, quantitative finance, energy system optimisation. These are the fields in which current AI systems already show capabilities that, on some benchmarks, exceed human level. Sam Altman has predicted that 2026 will see the arrival of systems capable of generating new scientific insights, moving from research assistants to autonomous researchers.

When it is expected

The largest survey ever conducted of AI researchers, with over 2,700 participants, estimates a 10% probability that AI systems will outperform humans in most tasks by 2027. There is no agreed date for the arrival of superintelligence. Forecasts vary because definitions, assumptions about technical progress and the criteria for measuring system capabilities differ. Researchers’ estimates and CEOs’ statements help in understanding possible scenarios, but they do not constitute a timetable for ASI.

Why it matters now

According to the 2025-2026 research by the Politecnico di Milano’s Digital Innovation in SMEs Observatory, 76% of Italian SMEs have neither invested nor plan to invest in artificial intelligence. The gap is cognitive. Those who are not building a working understanding of AI today will have to manage the impact of superintelligence without the tools to do so.

The debate dividing CEOs

The possibility of developing ever more capable systems has opened a debate on the speed of progress and on the time needed to understand its risks. The debate already concerns current AI and becomes even more relevant when the discussion turns to superintelligence.

Dario Amodei, CEO of Anthropic, has called for slowing the pace at which model capabilities increase. His proposal aims to give safety research, independent testing and institutions time to keep up. OpenAI has also argued that development must be able to slow down or stop when the available safeguards are insufficient.

In his speech to the UN on 22 September 2026, Trump instead stressed US technological leadership and rejected the idea of a global AI oversight system. His decision to call it super intelligence fits into this political debate: it presents the technology as a field in which to win and maintain American primacy. The name chosen by Trump refers to AI as a whole; the superintelligence described in this guide remains a future possibility.

For those who run a company, the debate has concrete consequences. The speed at which to adopt new systems, the suppliers to choose, the data to grant access to and the decisions to entrust to AI all call for different assessments depending on actual capabilities and risks. Preparing means building criteria for making these decisions and updating them as the technology changes.

Superintelligence and its impact on management: what really changes

Superintelligence is still in the future. Its precursors (agentic AI systems, models capable of autonomous reasoning, multi-agent architectures) are already here and are redefining managerial work.

According to a McKinsey estimate, by integrating available technologies, there is a technical potential to automate activities that take up 60–70% of working hours. This does not mean those activities are already automated: turning a technical possibility into a business process requires suitable tools, integration, skills and oversight. The impact also varies from one activity to another. AI systems can assist with synthesising information, processing documents and some repetitive processes; those who run the company must decide where to use them and which decisions to keep under human supervision.

Companies that McKinsey defines as “AI high performers” (those that deeply integrate AI into their decision-making processes) are 4.7 times more likely than their competitors to report a significant impact on EBITDA. This figure describes the present of applied AI. Superintelligence amplifies this gap in a non-linear way.

Leaders who want to prepare must answer three concrete questions. Which decision-making processes in my organisation can be augmented or automated with current systems? Which human skills must be protected and developed because they cannot be replaced? How do I build AI governance that stays fit for purpose even as systems become more capable?

The ability to answer these questions will determine the future of many organisations.

Prepare your company for the future of AI

Superintelligence is still in the future. Its impact on how brands are perceived, how markets take shape and how decisions are made is already here.

Companies that are building solid AI governance, a well-managed semantic narrative and a coherent brand structure across automated channels today will have a real competitive advantage as systems become more capable.

Language models already answer questions about products, make purchase recommendations and summarise corporate reputations. In the age of superintelligence this process will become faster still. A system with cognitive capabilities superior to those of humans will process every available signal about brands: every review, every article, every public action. It will then reach an unassailable assessment. A brand that does not take control of its own semantic narrative today is already losing ground in a market that will exist tomorrow, and that, when that day comes, will allow no appeal.