Summary
Aidan Gomez, CEO of Canadian AI company Cohere, makes a pointed case against the safety coordination proposals emerging from Anthropic and OpenAI, arguing that allowing a small number of frontier labs to set industry-wide safety standards amounts to creating an ‘AI cartel’ that entrenches their market position. Speaking on Bloomberg Tech, Gomez supports higher safety bars in principle but insists the process must include a much broader set of stakeholders and cannot rely on antitrust exemptions designed to prevent exactly this kind of coordination.
Gomez outlines four pillars he believes legitimate AI safety governance requires: an evidence-based risk framework with agreed-upon harm metrics, mandatory transparency and incident reporting requirements, capability testing scoped to high-risk domains within properly secured sandboxes, and assurance mechanisms built on broadly-developed rather than self-selected standards. He specifically critiques how the HuggingFace incident has been extrapolated beyond what the evidence supports, while affirming that the underlying capability testing is necessary — just not in insecure environments.
On Anthropic’s proposal to embed independent evaluators with employee-level access, Gomez expresses conditional support but flags a significant conflict-of-interest problem: the organizations being proposed for oversight — specifically Meter — have funding and personnel overlaps with Anthropic itself. He warns that a world where a small group self-regulates while simultaneously blocking competitive challenges to their leadership position is precisely the outcome antitrust law was designed to prevent.
📺 Source: Bloomberg Tech · Published September 14, 2026
🏷️ Format: Interview
