The chief executive of Toronto-based Cohere Inc. is publicly opposing a proposal by leading American artificial intelligence labs to coordinate on how quickly the technology advances, writing in an essay published Sept. 13 that the arrangement would amount to “a cartel by any other name.”
Aidan Gomez, who co-founded Cohere in Toronto in 2019 and co-authored the 2017 research paper that underpins modern large language models, aimed his criticism at a roadmap published this month by Anthropic chief executive Dario Amodei. According to Gomez’s essay, that roadmap asks governments for a narrow antitrust exemption so a small number of frontier labs can lawfully agree on shared safety standards and on limits to how fast the technology should progress, and asks that every other AI developer be required to follow whatever those participants settle on.
The point that has been largely missing from coverage of the essay, which has centred on the cartel comparison, is that Gomez is not arguing against regulation. “AI needs guardrails. That is not the dispute and never has been,” he wrote. “The dispute is over who writes them, who gets to participate and whose interests the rules are protecting.” He wrote that independent review of highly capable AI systems is a good idea and that Cohere supports it, and that his objection is to a designated list of companies existing at all when the resulting decisions would reach every government and citizen who was never consulted.
Gomez grounds the cartel comparison in two specific regulatory precedents. In 1975, he wrote, the U.S. Securities and Exchange Commission designated three firms as Nationally Recognized Statistical Rating Organizations and never published criteria by which anyone else could earn the designation. Twenty-five years later there were still only three, and they rated subprime mortgage securities triple-A. In 1985, he wrote, European carmakers won a sweeping antitrust waiver, the Motor Vehicle Block Exemption, on the argument that complex, safety-critical vehicles required manufacturer control over who was qualified to sell and service them. It took the European Commission roughly 25 years of reform to unwind. In both cases, Gomez wrote, the stated goal was safety and the result was a market structure that protected incumbents.
In place of a labs-led regime, the essay proposes four pillars: an evidence-based risk framework developed internationally and in the open rather than behind closed doors, mandatory transparency about how models are built and what risks they pose, independent testing scoped only to the capabilities that framework identifies as genuinely dangerous, and assurance mechanisms modelled on the licensing and inspection regimes used in aviation, nuclear power and financial services, in which evaluators are not paid by the parties they review. Rules should bind based on what a system can do rather than on who built it, Gomez wrote, so that a dangerous capability is treated the same whether it comes out of a large lab or a university department.
Separately, in an interview with CNBC’s The Tech Download published Monday, Gomez said advanced AI models are “the most potent cyber weapon that has ever been created,” citing their ability to find and exploit vulnerabilities at scale, and said companies should be deploying the same capability defensively to find weaknesses in their own systems first. He told CNBC he was not sure what a government oversight body would have done to prevent a July incident in which OpenAI said research agents circumvented network isolation controls during cybersecurity testing and reached systems belonging to Hugging Face and other third parties, according to reporting by the International Business Times. Anthropic disclosed separately that four of its Claude models gained unauthorized access to real third-party systems during capture-the-flag evaluations after misconfigurations left access to the real internet available.
The intervention lands during a week in which data sovereignty has been a recurring theme in Toronto, where the federal government is hosting Canada’s first investment summit. Gomez closed his essay on that argument, writing that hospitals, payment networks and defence ministries cannot pipe their most sensitive operational data to somebody else’s servers, and that security comes from sovereignty, local deployment and a competitive market with enough capable suppliers to absorb a failure at any one of them.
Anthropic has framed its proposal as a way to give developers time to complete safety work without sacrificing commercial advantage. Gomez’s essay quotes that framing directly and argues that a mechanism which slows everyone down while explicitly preserving existing commercial advantage does not make AI safer.
With reporting from the International Business Times, and from Aidan Gomez’s essay “Who Gets to Define the Rules for AI?”, published by Cohere on Sept. 13, 2026.
