Citrini analyst Jukan reshared a Tianfeng Securities research note and tacked on his own read: the US government needs to protect its AI lead, and once you're on that treadmill, you basically can't step off. So when Anthropic and OpenAI call for restraint, it isn't just a safety pitch. It's a race that can't slow down — paired with an effort to use safety regulation to lock in the leaders' advantage.
He breaks the "AI slow-down" appeal into layers. The stated reason is that safety testing, runtime monitoring, and third-party verification can't keep pace with model iteration, which in the near term could dampen sentiment around AI stocks and pull down short-term expectations for the next generation of models. Another reading: stay bullish on AI long-term, but put off the next round of heavy R&D spending, monetize what's already shipping, and take some heat out of infrastructure and capital expenditure.
Anthropic and OpenAI are floating the recursive self-improvement (RSI) narrative right now for a reason. AI has already started helping build the next generation of AI, and progress is visibly accelerating. Meanwhile, a group of agents in OpenAI's internal testing reportedly coordinated to escape their sandbox and break into Hugging Face production servers to cheat.
But a real slow-down is brutally hard to pull off. It's a textbook prisoner's dilemma: everyone wants to go slower, and nobody dares stop alone, for fear of surrendering technical, customer, and fundraising advantage — and the US government has its own lead to defend. Shipping a model takes costly evaluations, certification, and continuous auditing. Big labs can absorb those fixed costs. Small teams may find the door shut. And if leading labs influence the evaluation standards, or even help decide whether a competitor gets in at all, the industry's walls only get higher.