Citrini's view: Anthropic and OpenAI calling for an AI slowdown does not mean a real slowdown; the intent is to first monetize existing products and cool down capital expenditures.

Beating AI News Flash: Citrini analyst Jukan, citing a research report from Tianfeng Securities, expressed the view that the U.S. government needs to maintain its AI leadership. Once on this treadmill, it is basically impossible to get off. Anthropic and OpenAI's calls for deceleration cannot be seen merely as safety initiatives; behind them are multiple considerations: the race cannot slow down, while safety regulation is used to consolidate the advantages of leading players.

This further breaks down the multiple meanings of the "AI deceleration" call: the surface reason is that safety testing, operational monitoring, and third-party verification cannot keep up with the pace of model iteration, which in the near term may suppress sentiment in the AI sector and lower short-term expectations for next-generation models. Another interpretation is that they remain bullish on AI in the long term, but want to delay the next round of large-scale R&D, first monetize existing products, and at the same time cool down infrastructure spending and capital expenditure.

At this moment, Anthropic and OpenAI are putting forward the recursive self-improvement (RSI) narrative against the backdrop of AI already beginning to help develop next-generation AI, with progress clearly accelerating, and a group of Agents in OpenAI's internal testing reportedly collaborating on their own to escape the sandbox and intrude into Hugging Face's production servers to cheat. But deceleration is extremely difficult to achieve in reality. This is a typical prisoner's dilemma: everyone wants to slow down, but no one dares to stop unilaterally for fear of losing technological, customer, and financing advantages, while the U.S. government likewise must maintain its lead. Releasing models requires expensive evaluation, certification, and continuous auditing. Large companies can bear the fixed costs, while small teams may be shut out. If leading laboratories can influence evaluation standards or even participate in deciding whether competitors can enter, industry barriers will only become higher.