Aidan Gomez, CEO of Cohere, argues that major AI labs are using safety concerns as cover to lock in their market dominance through regulatory capture. He points out that Anthropic's recent proposal asks governments for antitrust exemptions so a handful of Silicon Valley companies can coordinate on AI safety standards and slow the technology's progress—while requiring all other developers to follow rules they had no voice in creating. Gomez sees this as a replay of past regulatory failures: the SEC's 1975 designation of three bond-rating firms that were never held to transparent standards, and Europe's 1985 Motor Vehicle Block Exemption that let car manufacturers control repair and service standards in the name of safety. In both cases, the stated goal was protecting the public, but the actual result was protecting incumbents and blocking competition for 25 years each. The core problem, Gomez argues, isn't whether AI needs guardrails—it does. The problem is who writes them and whose interests they serve. A safety regime designed by a small group of commercially aligned companies will only rigorously assess risks they've already built systems to handle, while ignoring other legitimate concerns. It will define risk purely in terms of model scale, automatically making only the companies with the largest systems "qualified" to judge safety. This conveniently entrenches their advantages. Gomez points out that real scientific disagreement exists about whether offensive capability comes from raw model size or from how the model is orchestrated with tools and verification steps—smaller well-designed systems can outperform massive ones in some tasks. A regulatory framework designed only around compute thresholds misses these nuances entirely. What makes the proposal particularly troubling to Gomez is a single sentence promising that coordinated safety work would give developers "time to do this safety work without sacrificing commercial advantage." He asks: whose advantage? The companies drafting the framework already sit at the market's top, so rules that slow everyone equally while explicitly preserving their current edge don't make AI safer—they calcify today's winners into position. Real safety standards should reduce actual risk regardless of who leads the market, not function as a mechanism to turn existing advantages into minimum requirements for competing at all.