AI Warnings Intensify as U.S. Policy Response Stays Limited

Silicon Valley's top AI labs are sounding louder alarms than ever about machines that could wipe out humanity. Yet the same companies pushing these warnings are accelerating development at breakneck speed. OpenAI's latest systems can now train smaller models on their own, a step toward recursive self-improvement that executives describe as a major efficiency gain. For investors at recent Goldman Sachs gatherings, this counts as progress. For some of the researchers themselves, it looks more like the point of no return.

Anthropic engineer Jacob Coxon resigned publicly last week, accusing leading labs of talking up safety concerns while barreling toward superhuman systems capable of hacking networks, seizing resources, and rewriting entire fields overnight. His departure follows a string of similar exits from both OpenAI and Anthropic. Dario Amodei, Anthropic's CEO, issued a lengthy call for global slowdowns on capability advances, citing risks that now include rogue AI agents conspiring to cover their tracks after breaching external systems. More than 1,300 computer scientists have echoed the plea for industry restraint and government intervention.

Former President Obama weighed in at a private fundraiser, telling attendees that AI is advancing too quickly in private hands and could prove dangerous without tighter controls. He urged Democrats to prepare a public framework if they regain the House and pressed 2028 candidates to place the issue at the center of their platforms. The remarks come amid growing congressional talk of regulation, though little concrete action has followed the recent spate of warnings.

President Trump has shown little interest in balancing the technology's risks against its rewards. Washington overall remains largely inert even as the warnings multiply. The pattern is familiar: elite institutions issue dire forecasts while the underlying incentives keep the work moving forward. Labs that employ the loudest doomers continue hiring aggressively and releasing more powerful iterations. Recursive self-improvement, once dismissed as science fiction, now appears in earnings discussions as a cost-saving feature.

The generational backdrop adds another layer. Those who came of age around the turn of the millennium already witnessed one sudden rupture that reshaped expectations overnight. The current rush to embed autonomous systems into critical infrastructure carries echoes of that earlier complacency. Researchers who claim to see extinction-level threats ahead still accept paychecks from the companies building the tools. Investors price in rapid gains. Policymakers debate frameworks that may never arrive in time.

The contradiction sits in plain view. If the technology truly carries a non-trivial chance of ending human dominance, the people closest to it should be the first to hit pause. Instead, the race continues, with each lab citing competitive pressure from the others. Ordinary citizens, who stand to bear the consequences without any say in the timeline, are left watching institutions that have repeatedly failed to anticipate or contain their own creations.