OpenAI, Anthropic, and Google have publicly advocated for slowing frontier AI development, framing the proposal as a safety imperative. The three giants argue that measured advancement allows time for proper alignment research and risk mitigation before deploying more powerful systems. Yet this unified stance faces fierce resistance from competitors, policymakers, and industry observers who view the slowdown as a calculated business strategy rather than genuine safety concern.
Cohere CEO Aidan Gomez has been the most direct critic, labeling the initiative a "cartel by another name" designed to entrench market dominance by smaller players. His accusation carries weight in an industry where first-mover advantage translates to data, compute, and talent accumulation. A slowdown imposed by the three largest AI labs would effectively freeze the competitive landscape at its current position, making it harder for well-funded startups and international competitors to catch up.
The Trump administration and White House have also opposed the slowdown, signaling that the U.S. government views rapid AI advancement as strategically necessary. This position reflects geopolitical anxiety about China's AI capabilities and the desire to maintain American technological leadership. Slowing development domestically while competitors abroad accelerate creates national security risks that administration officials take seriously.
Critics question the timing and motivation behind the safety narrative. These three companies have collectively invested billions in frontier model training and deployment. Their proposed slowdown comes after they have already trained and released increasingly capable systems. Skeptics argue this resembles closing the barn door after the horses escape. Anthropic's focus on Constitutional AI and safety messaging, paired with its simultaneous scaling of compute and model parameters, creates a tension that observers find unconvincing.
The proposal also conveniently insulates incumbents from disruption. Startups with novel architectures or training approaches need rapid iteration cycles to prove their methods work at scale. A mandated slowdown disadvantages these challengers while allowing established players to consolidate their positions through software, services, and integration strategies. Companies like xAI, Meta, and Mistral have demonstrated they can produce competitive models outside this "safety coalition," suggesting the restriction targets competition as much as risk mitigation.
Industry voices from research labs and smaller companies have pointed out that safety advances need not require slower development timelines. Better interpretability, testing frameworks, and evaluation methods can scale alongside model capabilities. The proposal conflates speed with recklessness, but these are distinct variables. A company can advance quickly and safely, or slowly and recklessly.
The White House opposition particularly damages the safety framing. Administration officials publicly stated that slowing AI development threatens American competitiveness and economic growth. This stance prioritizes innovation over caution at the highest policy level, undercutting the moral authority of the safety argument.
What remains unclear is whether this disagreement stems from genuine disagreement about safety or from standard competitive positioning dressed in safety language. Both can be true simultaneously. These companies may believe in safety while also recognizing that a coordinated slowdown benefits them strategically. The public cannot distinguish between authentic caution and self-interested gatekeeping based on rhetoric alone. The persistence of rapid development across the industry suggests that most participants, including within OpenAI, Anthropic, and Google, prioritize capability advancement over the proposed restraint.