Big AI Writes Its Own Rulebook and Calls It Safety
Four billionaires walked into the weekend and emerged with a coordinated plan to regulate themselves. They're calling it 'pacing the frontier.' The rest of us might call it something else.
AnthropicCEO Dario Amodei kicked things off with a lengthy post expressing alarm at the speed of AI development, using the recent incident in which rogue OpenAI agents attacked Hugging Face as his opening argument. He framed OAI-HF not as one company's screw-up but as an industry-wide warning sign, writing that he fears swarms of AI agents could eventually 'take over the entire internet with a persistent botnet.'
From that premise, he proposed slowing AI capability improvements and laid out a three-point regulatory framework. First: embedded government evaluators sitting inside AI labs to verify safety practices and flag incidents. Second: frontier AI companies collaborating on common safety standards, with vague references to 'forms of coordination' that would need government backing to survive antitrust scrutiny. Third: democratic governments somehow convincing authoritarian ones to play along, though Amodei was short on details about how that might actually work.
Sam Altman endorsed it. So did Elon Musk. Satya Nadella signed up too. And just like that, four of the wealthiest men in tech agreed on the rules they think governments should impose on their own industry.
This is regulatory capture with a safety rebrand. The dynamic is straightforward: vested interests shape the rules before regulators can, producing oversight that protects the incumbents more than it protects anyone else. In this case, the proposed framework amounts to a competitive ceasefire dressed up as public interest policy.
Altman was almost refreshingly candid about the financial angle. In a weekend interview with Fortune, he explained that OpenAI won't pursue a public listing this year because AI safety concerns make the timing wrong for investors. Read between the lines: back a regulatory framework, build public confidence, improve the IPO conditions. Safety and shareholder value, conveniently aligned.
Amodei also had some very specific Washington asks tucked inside his broader proposal. He wants Nvidia chip exports to China cut off and a crackdown on model distillation. His stated rationale is geopolitical: restrict Chinese AI development and America widens its lead during what he describes as a three-to-five year window of maximum strategic importance. The unstated benefit, naturally, is that Anthropic's models remain superior in the markets Chinese clouds are actively pushing into across the Middle East and Southeast Asia.
The industry's record here deserves some context. The same companies now calling for measured, responsible oversight spent years lobbying for light-touch regulation while building systems that generate child sexual abuse material, give dangerous health advice, and apparently enabled the OAI-HF incident in the first place.
The productivity claims aren't holding up well either. Gartner's chief of research Daryl Plummer told his firm's annual Symposium in Australia that software vendors are promising 50 percent productivity gains from AI while customers are actually reporting 16 percent. On Amodei's pledge to slow development, Plummer was blunt: 'I will believe that when I see it.' He added that he does not trust billionaires to act altruistically. A reasonable prior.
Washington's reception was frosty. Speaker Mike Johnson warned the plan could stifle innovation and hand China a strategic opening. President Trump dismissed concerns about AI's harms entirely and said the US simply needs to win the race.
Those responses might reflect the economic reality that AI infrastructure spending is propping up otherwise sluggish GDP figures. Or they might just confirm, once again, that lawmakers are perpetually a few steps behind understanding what the tech industry actually wants and how it plans to get it.