Aidan Gomez, CEO of the Canadian firm Cohere, raised a point that deserves sustained attention. A small cluster of Silicon Valley companies shouldn't be the sole custodians of AI safety. He went further still, comparing the apocalyptic warnings coming from American firms to a wolf dressed in sheep's clothing. That image captures with precision a trend that has been building for some time.
AI governance refers to the set of rules, institutions, and accountability processes that determine who decides which capabilities an AI system is allowed to have. Without that framework, any safety pledge collapses into mere corporate messaging. Gomez pinpoints exactly this weakness. When the people building the models are also the ones defining what counts as risk, safety stops being a commitment and becomes a strategy.
Cohere competes head-to-head with OpenAI, Anthropic, and Google DeepMind. Even so, its CEO publicly pointed out that his competitors' alarmist rhetoric functions as a regulatory moat. When a company claims its technology is so dangerous that only it can oversee it, a question naturally follows: who actually benefits from that narrative?
Almost always, the company repeating it. Calibrated fear builds walls that others can scarcely climb.
The Generosity in the Doorway examines the recent AI ecosystem and surfaces a recurring thread. Sam Altman writes about the singularity while negotiating with governments and defining existential risks on his own terms. Yann LeCun has challenged that same alarmism from within Meta. Jensen Huang has redrawn the concept of AGI to suit Nvidia's interests. Whoever controls the technical vocabulary tends to dictate the terms of the regulatory debate as well.
The Gomez case adds a geographic dimension that shifts the weight of the argument. This criticism isn't coming from a European regulator or an academic. It's coming from Canada, from another player within the industry itself. He isn't calling for innovation to be slowed down. He's questioning whether precaution, defined unilaterally by a handful of California companies, is in fact a form of market dominance in disguise.
There are documented examples of failed governance that support this reading. Anthropic leaked half a million lines of code and blamed it on human error. The Pentagon flagged it as a supply-chain risk. A federal judge determined it functioned more as retaliation than as a technical assessment. Concentrated power. Decisions made without external checks.
What does this mean for someone who doesn't work in tech or follow these debates closely? That every time a major firm calls for more regulation in urgent tones, it's worth examining who drafted the proposal and who stands to gain from the threshold being set. The history of industrial regulation is full of cases where dominant players pushed for strict rules that only they could afford to meet. It hardly seems a coincidence that the largest generative AI companies are also the most insistent on controls they already satisfy or can easily bear the cost of.
This confirms that tech self-regulation tends to fail, though with one caveat. There are researchers within these organizations who genuinely believe in the risks they describe. It's the structure that determines the outcome. Without independent external audits and accountability mechanisms beyond their reach, the result tends to repeat itself regardless of individual intentions.
I recognize these patterns from other contexts. The most effective checks rarely emerge from within. They require external tension, genuine competition, and overseers who don't depend on funding from the very entities they regulate. Gomez, without fully intending to, is pointing toward exactly that need.
One question remains open, and neither Gomez nor I resolve it here: who has both the technical capacity and the political independence to build that external framework without it eventually being captured by the very interests it's meant to watch over?
