Cohere’s Nick Frosst, the AGI sceptic at the heart of a $20bn sovereignty bet

Nick Frosst has picked a loud week to make a quiet argument. As Cohere signed its definitive merger with Germany's Aleph Alpha, a combination valued at roughly $20bn and pitched as the first transatlantic sovereign AI provider, the Toronto company's co-founder sat for a long interview and argued that the destination the entire industry is pricing in, artificial general intelligence, is not on the map. The sharpest edge is personal. Frosst learned the field from Geoffrey Hinton, who hired him as one of his first researchers at Google Brain in Toronto in 2016, and still calls him a dear friend. He also thinks the great man's timeline is wrong, and he said so in print.

Cohere Watch analysis

What happened

The interview, published on 17 September, ranges from enterprise procurement to indie rock, but its spine is that disagreement. Hinton has spent recent years warning that the technology may carry us towards something genuinely human-like and genuinely dangerous. Frosst, who co-founded Cohere with Aidan Gomez and Ivan Zhang in 2019, calls the warnings worth having and the timeline wrong. "I do not think the technology we have will get us to AGI anytime soon," he said. In his telling, large language models are a transformative tool that is distinctly not human-like, and making them better pushes them towards what they already do well, not towards a general mind.

The interview's most striking passage concerns June. It recounts a Friday letter from the US Commerce Department to Anthropic, citing national security, barring foreign nationals from the company's newest systems and forcing two frontier models, named as Fable 5 and Mythos 5, offline for every user at once. The controls were lifted on 30 June. Frosst said the episode made tangible what Cohere had argued for years: if foundational technology runs on another jurisdiction's permission, someone can switch it off with an afternoon's notice. Joelle Pineau, Cohere's chief AI officer, reported a huge number of inbounds from companies and governments once it happened.

Beneath both threads sits a definition of sovereignty that Frosst would like to rescue from the geopoliticians.

Why it matters

The safety argument of recent weeks has run almost entirely one way: a viral resignation, a slowdown essay, warnings from the field's most cited names. Frosst is a frontier insider arguing that the premise itself, that scaling leads towards an approaching general intelligence, is mistaken. If he is right, policy and capital aimed at imminent AGI will misfire, and the interesting questions move to deployment and control. If he is wrong, a $20bn sovereign AI pitch is priced on a map with the destination missing. Either way, the disagreement is now explicit, public and testable.

The June story is the argument's load-bearing anecdote, and it deserves care. It arrives as one interview's account of a government action, striking enough that corroboration should be wanted before it hardens into history. But the commercial point does not depend on every detail: the buyers Frosst describes, governments and regulated industries, are paying precisely so that a single letter cannot darken their systems.

Our read

Frosst is worth reading because he sells neither doom nor utopia. His labour argument is the practical version of his scepticism: technology that automates some percentage of everybody's job rather than a full percentage of anybody's, with programming as the exhibit, tools that took the easy parts while the industry keeps hiring engineers because architecture and judgement still need a person. Treat the AGI position as a considered bet from someone with standing, not a benchmark, and note that it is falsifiable. Read the Observer interview for the worldview in full, kiln-tending home model and touring band included, then keep half an eye on whether the June directive gets the second source a claim that size deserves.

What to watch

• The June directive
  Whether further reporting or documents confirm the Commerce order and the Anthropic outage as the interview describes it.
• The merger's passage
  Closing is expected later this year, subject to regulatory approvals.
• The Series E
  Schwarz Group's commitment and the final shape of the round.
• Scepticism in practice
  Whether Frosst's AGI position shapes Cohere's research and product claims or stays confined to interviews.

Discussion spark: Frosst bets that scaling makes better tools rather than an approaching mind, and he is a co-founder of the $20bn bet built on that view. Is his AGI scepticism the clearest reading in the industry, or exactly the belief a sovereignty vendor needs its customers to hold?

not affiliated with or endorsed by Cohere