10 AI and Superintelligence Voices Worth Following in 2026

Anyone who follows AI soon finds that the loudest voices are rarely the most informative, and that the experts disagree about basic things. Some argue that systems far smarter than people are close and need urgent safeguards. Others call the idea of superintelligence incoherent, or argue that today's systems are overhyped. Both camps include serious researchers.

This list does not try to settle that. It names ten people whose public work is worth reading, chosen so that they disagree, and for each one it gives a role we checked on a primary page, a link to their own words, where they differ from others and how to read them with care.

1. Yoshua Bengio: a researcher who says safety-minded people should leave frontier labs

Role, as stated by the source: His own site lists him as Full Professor of Computer Science at Université de Montréal, Co-President and Scientific Director of LawZero, Founder and Scientific Advisor of Mila, and chair of the International AI Safety Report.

Known for: Writing about AI risk on his site. On 8 October 2026 he published "If you prioritize safety, leave frontier AI companies", an op-ed in which he says his own deep learning research helped bring about current risks and argues that people who put safety first should leave frontier AI companies.

Where they differ from others: His call to leave frontier companies contrasts with the position of the company leaders on this list, who work inside such companies, though we found no source in which they respond to each other. We read only the excerpt his site republishes, not the full op-ed.

2. Geoffrey Hinton: a Nobel laureate who warns about control

Role, as stated by the source: The University of Toronto describes him as University Professor Emeritus of computer science and the 2024 Nobel Prize in Physics recipient. His page says he worked half-time at Google from 2013 to 2023, ending as a Vice President and Engineering Fellow.

Known for: Neural-network research and, more recently, public warnings. In a University of Toronto report from 27 June 2025, he is quoted as warning that AI could pose an existential threat through misuse by bad actors and through super-intelligent systems acting outside human control, and as arguing that large language models understand language. His own page is here.

Where they differ from others: He argues that language models understand language and that regulation is needed. Whether models "understand" is contested, but we did not read a source in which he and a named critic debate it, so we pair him with no one.

3. Stuart Russell: the control problem, set out for general readers

Role, as stated by the source: His page lists him as Distinguished Professor of Computer Science at UC Berkeley and Smith-Zadeh Professor in Engineering, with affiliations including the Center for Human-Compatible AI.

Known for: His book *Human Compatible: AI and the Problem of Control*, described on his page through reviewers' blurbs as an account of how artificial super-intelligence threatens human control and of a new human-computer relationship in response. His page also lists the 2021 Reith Lectures, Living With Artificial Intelligence.

Where they differ from others: His book is described as treating control of advanced AI as a problem that needs a new solution. The 'normal technology' authors below call the superintelligence framing incoherent. That contrast comes from comparing their texts, not from a source in which either names the other.

4. Dario Amodei: a company leader writing about the risks of what his company builds

Role, as stated by the source: His site describes him as the CEO of Anthropic, which it calls a public benefit corporation dedicated to building AI systems that are steerable, interpretable and safe.

Known for: Essays on AI's promise and risks. "The Adolescence of Technology" (January 2026) describes powerful AI as a "country of geniuses in a datacenter" that could arrive within a few years, and names four main risks: autonomy risks, misuse for destruction, misuse for seizing power and economic disruption. It says the risks can be overcome with careful action and calls for evidence-based responses, listing remedies such as transparency legislation and limits on chip sales to China.

Where they differ from others: He expects powerful AI within a few years and lists company practices such as audits and system cards among remedies, alongside legislation. Skeptics on this list question the timeline and the framing, and Bengio's op-ed urges safety-focused people to leave frontier companies. We found no source in which either responds to the other.

5. Demis Hassabis: the case for AGI as a scientific tool

Role, as stated by the source: A Google post says he is moving from day-to-day operations at Google DeepMind to a strategic role as chair of GDM and Alphabet's top scientist, and that he keeps leading Isomorphic Labs. The post carries no date. Google DeepMind's about page still lists him as co-founder and CEO, so the two pages conflict; we followed the announcement.

Known for: Leading research towards AGI. In the Google message he writes, "I've been working towards AGI my whole life and now, like many of you, I feel it is close at hand." In a 10 March 2026 essay he writes that "AGI will be the most profound technology ever invented" and that specialised AI tool use will be critical for reaching it.

Where they differ from others: His posts emphasise AGI's potential for science, medicine and productivity and say it is close, where the safety-focused researchers above emphasise risks. His posts do not respond to them.

6. Gary Marcus: a persistent skeptic of current approaches

Role, as stated by the source: His newsletter shows only the byline "Gary Marcus" and a subscriber count above 121,000. It does not state his role or affiliation, so we give none.

Known for: Arguing that current AI is unreliable and overhyped. In "We must recall open-ended AI agents with internet access from the market, now" (10 October 2026) he argues such agents should be pulled until their defects are fixed, writing that "the current generation of synthetic agents simply cannot be trusted." Other October posts include "Generative AI in disarray" and "Could this be the Frontier Labs' WeWork moment?"

Where they differ from others: He wants agents recalled and doubts the finances and safety of frontier labs, where company leaders and many users take the opposite view.

7. Arvind Narayanan: AI as a normal technology

Role, as stated by the source: His page lists him as Professor of Computer Science at Princeton University and Director of its Center for Information Technology Policy.

Known for: With Sayash Kapoor, the essay "AI as Normal Technology" (15 April 2025) and the newsletter at normaltech.ai. They write, "We view AI as a tool that we can and should remain in control of," and expect impacts to unfold over decades. His page also lists the book AI Snake Oil.

Where they differ from others: The essay describes "superintelligent" AI as "incoherent as usually conceptualized" and places most risk in misuse and systemic harms, not in a loss of control. That is a direct challenge to the framing of Hinton, Russell and others above.

8. Melanie Mitchell: a check on hype from a complexity scientist

Role, as stated by the source: The Santa Fe Institute lists her as James B. Alley, Jr. Professor and a member of the Science Steering Committee, with Portland State University named as her home institution. Her research focus is given as "conceptual abstraction, analogy-making, and visual recognition in artificial intelligence systems."

Known for: Her book *Artificial Intelligence: A Guide for Thinking Humans* (Farrar, Straus and Giroux, 2019), which its page describes as exploring "the profound disconnect between the hype and the actual achievements in AI." Her Santa Fe Institute profile links to her home page and publications.

Where they differ from others: She stresses the gap between claims and demonstrated ability. The pages we read name no specific person she disagrees with, so we do not name one.

9. Timnit Gebru: a critic focused on who benefits

Role, as stated by the source: The DAIR site lists her as founder and executive director of DAIR, whose own words are: "We believe that research should center the people most impacted by technology and should be rooted in community needs."

Known for: Advocacy for community-centred technology. Her page promotes her forthcoming memoir, Deep Unlearning, which it says argues for "a technological future that serves our communities" and contrasts that with AI used for surveillance, warfare and concentrating power in Silicon Valley.

Where they differ from others: She focuses on present-day harms and the concentration of power, a different emphasis from the safety researchers above, who concentrate on loss of control. The pages we read do not set her view against any named person.

10. Ethan Mollick: a practical guide to using AI at work

Role, as stated by the source: His Wharton profile lists him as Associate Professor of Management, Ralph J. Roberts Distinguished Faculty Scholar, and Co-Director of Generative AI Labs at Wharton.

Known for: Practical experiments with AI at work and in school, through his newsletter One Useful Thing, described as "trying to understand the implications of AI for work, education, and life," and his 2024 book Co-Intelligence: Living and Working with AI (Penguin Portfolio), both listed on his Wharton profile.

Where they differ from others: He is mainly a guide to use, not a participant in the superintelligence debate, so we found no source setting him against another voice on this list.

How should I follow people who disagree?

Where can I read more on this site?

• AI alignment and superintelligence risks explain the problems the safety researchers describe.
• Superintelligence timelines collects published forecasts and their limits.
• AI safety incidents and warning signs of 2026 covers documented incidents that skeptics and safety researchers both cite.
• 10 companies shaping AI covers the organisations several of these people lead.
• Vendor claims vs. verified findings explains how we label what a speaker says against what has been checked.

What are the limitations of this list?

• Ten people cannot represent a field. The list leans towards English-language researchers and public writers, and omits policy officials and journalists because we could not read primary pages that set out a position for them.
• We read each source on 18 October 2026, and several are long. Where we read only part of a page, we say so in the entry.
• Roles change. Hassabis's role is described differently on two Google pages, and others may have moved since the pages we read.
• We describe positions as sources state them and have not verified the underlying claims about incidents, finances or timelines.