The fight over who writes AI safety rules has moved from conference panels to antitrust law, and a Canadian company just took the other side.
- Cohere CEO Aidan Gomez published an essay on 13 September 2026 calling a coordinated AI slowdown a cartel by any other name.
- The objection is specific: the proposal asks governments for a narrow antitrust waiver so rival labs can lawfully coordinate.
- Whoever wins this argument sets the compliance cost that lands on every Canadian company buying AI.
The AI slowdown argument reached the point this month where somebody asked a government for permission to collude. Cohere co-founder and CEO Aidan Gomez answered on 13 September 2026 with an essay arguing that a handful of market dominant Silicon Valley firms should not get to write the rules for everyone else. His word for the proposal was cartel. He is a Toronto CEO with a commercial interest in the outcome, and the argument still deserves a hearing on its merits.
TL;DR
- Gomez published Who Gets to Define the Rules for AI on 13 September 2026.
- He objects to the request for a narrow antitrust waiver that would let rival labs coordinate on pacing and standards.
- His argument is that coordination preserves existing commercial advantage while slowing everyone down.
- He proposes four alternatives: evidence based risk frameworks, mandatory transparency, independent testing scoped to dangerous capabilities, and assurance modelled on aviation and nuclear.
- He warns against compliance frameworks that choke off smaller labs before they ship anything.
What is the AI slowdown proposal
Several frontier labs have floated a pacing framework, a coordinated agreement to slow the release of the most capable models while safety testing catches up. The mechanism is the problem. Competitors agreeing on what they will and will not ship is the textbook shape of a restraint of trade, which is why the proposal comes attached to a request for a narrow antitrust waiver.
Gomez wrote that the plan preserves existing commercial advantage while slowing everyone down. The Globe and Mail reported the same week that he is not arguing against testing. He wants independent testing scoped to capabilities that are actually dangerous, transparency obligations that apply to everyone, and an assurance regime built the way aviation and nuclear built theirs.
That is a narrower disagreement than the headline suggests. Almost nobody in this fight is against safety testing. The fight is about who runs it.
Why it matters in Canada
Canada has one company in the frontier conversation and it is objecting to the governance model. That is worth noticing on its own.
The practical stake is downstream, and it is yours. If a small group of labs sets the pacing rules, those rules become the de facto compliance baseline that flows into vendor contracts, procurement questionnaires and insurance forms. You will not attend the meeting. You will receive the invoice.
There is a second effect. Gomez warns about frameworks that choke off smaller labs before they ever ship. A Canadian business shopping for AI in 2029 wants a market with more than three suppliers in it, because three suppliers is how you get roadmap pricing and take-it-or-leave-it terms. Buyer choice is a safety property too, just not the kind that gets a white paper.
What does this cost a business that is not a frontier lab
Nothing today. Quite a lot by the time it reaches you.
Compare the two paths. Under coordinated pacing, the rules arrive as vendor terms, and your compliance work is whatever your supplier passes down. Under an assurance regime of the kind Gomez describes, the rules arrive as published standards you can read, test against and switch supplier over. The first is cheaper to start and more expensive to leave. The second front-loads the work and keeps the exit open.
Canadian firms are not spectators in this, even at 40 employees. Statistics Canada reported on 31 August 2026 that 25.2% of businesses plan to use AI over the next 12 months. The governance model being argued about now is the one that will be written into those contracts.
What should leaders do next
- Ask every AI vendor which external standard they test against and who publishes it. If the answer is an internal policy, write that down.
- Keep a second supplier viable for any AI workload you would struggle to replace in a quarter.
- Read your AI clauses for pass-through compliance language. The phrase to find is any obligation that changes when the vendor adopts a new framework.
- Do not wait for a Canadian federal AI statute to set your internal rules. Ottawa has not moved fast, and your exposure is contractual either way.
- If you belong to an industry association, push it to comment on AI governance consultations. Buyer voices are almost entirely absent from this argument.
The AI Podcast covers how these governance fights land on Canadian operators, and the Alberta AI Advantage sets out what a province can do while the federal picture stays unsettled.
Why the other side might be right
Steel-man it properly, because the pacing argument is not silly.
If a capability jump genuinely creates catastrophic risk, a competitive race to ship it first is the worst possible governance mechanism, and antitrust law was not written with that case in mind. Aviation safety works partly because manufacturers do coordinate, through standards bodies and regulators, on things nobody competes over. A time-limited waiver with a regulator in the room is not obviously a cartel.
And Gomez runs a company that competes with the labs proposing the pause. That does not make him wrong. It does mean his position and his commercial interest point the same direction, which is worth saying out loud.
What to watch over the next 90 days
- Whether any competition authority in the United States, the European Union or Canada responds publicly to the antitrust waiver request before year end.
- Whether other mid-sized labs join Cohere publicly. One objector is a position, four is a bloc.
- Canadian federal signals on AI legislation through the fall sitting, which will decide whether Ottawa writes rules or inherits them.
FAQ
What is the proposed AI slowdown?
A coordinated pacing framework under which leading labs would agree to slow releases of the most capable models while safety testing matures, with a narrow antitrust waiver to make the coordination lawful.
Why does Cohere call it a cartel?
Because competitors would agree among themselves on what to ship and when. Gomez argues that preserves the market position of the firms already ahead.
What does Gomez propose instead?
Evidence based risk frameworks built transparently across countries, mandatory transparency for developers, independent testing scoped to genuinely dangerous capabilities, and an assurance regime modelled on aviation, nuclear and finance.
Does any of this affect a Canadian small business today?
Not directly. It affects the contract terms and compliance questions your AI suppliers will pass to you within two to three years.
Closing analysis
Canadian companies have spent two years being told AI governance is something that happens to them. This argument is the opening where that changes. The rules being drafted now will arrive in your vendor contracts long before they arrive in legislation, and the buyers who read those clauses early are the ones who keep a choice of supplier.
Sources
- Aidan Gomez, Cohere, Who Gets to Define the Rules for AI, 13 September 2026. cohere.com
- The Globe and Mail, AI firms calls for co-ordinated slowdown amounts to cartel behaviour, Cohere CEO says, September 2026. theglobeandmail.com
- Statistics Canada, Canadian Survey on Business Conditions, third quarter 2026, 31 August 2026. statcan.gc.ca
- Bank of Canada, Canadian businesses use of AI, what the evidence shows, August 2026. bankofcanada.ca
Related reading
- AI Transparency in Canada Needs a Practical Company Record
- AI Tool Fatigue and the Business AI Switchboard
- Meta Muse Glimmer and Local AI for Canadian Business
Disclosure
The author has no relevant financial, advisory, or board relationships with any party named in this column.
Zak Hussein writes on AI for Canadian business owners and operators. He is the founder of AI Magazine Canada and CEO of ORKA AI.
Part of our continuing reference on Canadian AI policy.
