AI boom or doom: what will superintelligence feel like and when will we get there?

Deutsche Bank Research Institute

AI boom or doom: what will superintelligence feel like and when will we get there? October 12, 2026 The recent panic about rogue AI agents collaborating in swarms to take over the internet and threaten human existence has, more than occasionally, lacked Authors nuance. But it does reflect how artificial intelligence is progressing in creative Adrian Cox and surprising ways, and at a pace that seems uncontrollable. Research Analyst

For some in the industry, it is evidence that the technology is following a path of Azeem Azhar, Exponential View accelerating self-improvement that will lead to artificial general intelligence (AGI) Expert Collaborator and, perhaps, to utopia or catastrophe. Former Google CEO Eric Schmidt has sketched out a “San Francisco Consensus” based on a belief in scaling laws, with ever larger AI models becoming an input into their own improvement; superintelligence by the end of the decade; and transformative AI systems outperforming humans on many tasks. This loop would lead to exponential growth in scientific advancement, financial returns and human progress. And crucially, AGI, and then superintelligence, would merely be milestones in a process, far from its climax. If that were the case, superintelligence could make much of the cognitive work that dominates the world economy much cheaper and change the ways in which firms compete. But its economic effects will depend on how businesses and customers respond. In this paper, we analyse what the San Francisco Consensus means for investors and enterprises in the years to come and what could accelerate – or slow down – its timeline of AGI by 2030. 1. What would superintelligence look and feel like? 2. Can we get there theoretically? 3. Can we get there practically? 4. What happens if we don’t get there? The base case The question is urgent. And it is complicated by reports of AI models autonomously accessing the internet and hacking other companies; a public backlash over the energy- intensive data centres springing up as part of a historic $1 trillion a year investment boom; and dire warnings of threats to life if superintelligent AI runs roughshod over its human creators to achieve its own goals.

Important research disclosures located in Appendix 1 1

12 October 2026 Thematic Research

1. What would superintelligence look and feel like? 1.1 What are AGI and superintelligence? Artificial general intelligence (AGI) is a simple milestone scribbled in pencil: an AI system that has surpassed human capabilities. Superintelligence is the continuum beyond that point. It is not merely systems that exceed the best human performance across one cognitive task or another, but systems that exceed human potential across every cognitive task.

The San Francisco Consensus takes this further: the intelligence must be able to produce more intelligence. The key to getting there is so-called recursive self- improvement, where the AI is able to autonomously design and improve itself in a way that compounds exponentially. A first step has maybe already been taken, with OpenAI saying recently that AI is now writing approximately 80 percent of the company’s code.

There is significant disagreement on how we will be able to identify AGI or superintelligence when we see it, let alone what comes afterwards.

OpenAI’s Charter defines AGI in economic terms, as “highly autonomous systems that outperform humans at the most economically valuable work”. Google researchers proposed a five-step ladder based on performance and generality, with level 5 “superhuman” for a system that outperforms 100% of humans. And philosopher Nick Bostrom, who got the ball rolling in 2014 with his influential book “Superintelligence”, defined it as “any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest”.

In August, OpenAI CEO Sam Altman said the company would have an internal system that he would call AGI by later this year but subsequently told a podcast that “at best it’s a very poorly defined term. I was going to say it’s like an irrelevant marketing term”. Yet his No. 2 Greg Brockman described the new Astra model at the start of…