The AI safety debate shifts from public declarations to a private standards body

The accelerator and the brake in the same room

The tension between calls to slow development and the desire to avoid government regulation reached a boiling point this week. While Anthropic chief executive Dario Amodei published a roughly 4,000-word essay demanding coordinated international slowdown and guardrails before releasing more powerful models, the largest rivals — OpenAI's Sam Altman and xAI's Elon Musk — rushed to endorse it publicly. At the same time, those same companies are now setting up a private standards body that would let them write the rules themselves.

Amodei's plan: global cooperation instead of a race

According to Amodei, capabilities are advancing faster than the ability to secure them, so an international strategy combining companies and governments is needed for safe deployment. The backing of Altman and Musk gives the effort political weight, but it also exposes the paradox: the labs pushing for a slowdown are the ones profiting from the race, and they will be the ones defining what "slowdown" looks like in practice. The document contains no quantitative targets or binding timelines, only a principled call for coordination.

Zuckerberg: the market will sort it, not government

Mark Zuckerberg staked out the opposite position on X. Trust and alignment, he argued, are becoming the differentiating capabilities between agents and models, and Meta has already delayed the launch of Muse by months to invest in safety and security — without demanding competitors do the same. "We just did it as part of the ongoing work because it was the right thing for people and for us," he wrote. The implicit message: organic market incentives will force everyone into line, and government regulation is unnecessary. The stance aligns with the essay he published last month, which emphasized American competitiveness and only limited cooperation with authorities.

Other voices: "tone-deaf" and capabilities outpacing safety

Reddit co-founder Alexis Ohanian told CNBC the industry had been "tone-deaf" in explaining risks to the public, yet expressed hope the companies would solve the problems themselves. Google DeepMind co-founder Shane Legg warned that capabilities are advancing "very, very fast" and must not be allowed to outpace safety: "We need to go through the details and see how it works in practice." Neither offered a concrete enforcement mechanism.

A self-regulatory body instead of a federal one

According to a report from The Information, OpenAI, Anthropic and other large companies are jointly establishing a private standards organization — a kind of self-regulatory body. Discussions about a parallel federal entity have taken place in the past, but the Trump White House, which includes vocal AI supporters led by "AI czar" David Sacks, advanced nothing. Sacks declared that regulation should remain in the companies' own hands. The result: the industry writes the rules that will govern it, while asking the public to trust that it will slow down when it decides doing so is worthwhile.