Analyst: US AI Race Difficult to "Slow Down," Safety Regulation May Instead Entrench Leading Labs' Advantage

Odaily News: Citrini analyst Jukan shared a research report from Tianfeng Securities and stated that the US government needs to maintain its leading position in AI, so once it enters the AI race, it is very difficult to truly stop. Jukan believes that recent calls by Anthropic and OpenAI to slow down AI development cannot be viewed merely as safety initiatives; behind them may simultaneously lie multiple considerations, including the inability to slow the competition and the desire to entrench leading advantages through safety regulation.

Jukan further pointed out that the relevant "AI slowdown" calls ostensibly stem from safety testing, operational monitoring, and third-party verification being unable to keep pace with model iteration speed, which in the short term may suppress market sentiment in the AI sector and lower market expectations for next-generation models. Another possibility is that the industry remains bullish on AI in the long term but wants to postpone the next round of large-scale R&D investment, prioritize commercializing existing products, and reduce pressure on infrastructure and capital expenditure. He believes the AI race is essentially akin to a "prisoner's dilemma" — all parties want to slow down, but none dares to be the first to stop, lest they lose their technological, customer, and financing advantages.

Jukan also noted that Anthropic and OpenAI's recent emphasis on recursive self-improvement (RSI) is related to AI having already begun assisting in the development of next-generation AI, accelerating model iteration speed. Meanwhile, in OpenAI's internal testing, incidents reportedly occurred in which Agents collaborated to escape the sandbox and intrude into Hugging Face production servers. Jukan believes that as model releases require bearing expensive evaluation, certification, and continuous auditing costs, large labs are better positioned to absorb these fixed costs, while smaller teams may face higher barriers to entry as a result. If leading labs further participate in setting evaluation standards, industry barriers may continue to rise.