AI lab leaders talk about slowing down, but the details are missing

A public debate flared this week over whether the AI industry is actually prepared to tap the brakes. Dario Amodei, Anthropic's chief executive, published a plan called "Pace the Frontier" that calls for a controlled slowdown in advanced model development. Meanwhile, Nvidia chief Jensen Huang echoed President Trump's claim that the public backlash against AI is a hoax and that regulation is unnecessary. On the latest episode of TechCrunch's Equity podcast, Anthony Ha, Sean O'Kane and Kirsten Korosec tried to figure out whether CEOs like Amodei and OpenAI's Sam Altman are serious.

O'Kane admitted his conviction from the previous week — that no slowdown stood a chance — flipped after hearing Amodei, Altman and, to a lesser extent, Elon Musk discuss the topic. He remains skeptical, though. Amodei's plan, laid out in a blog post, is thin on detail. Missing are not only concrete arguments for the dangers they flag, but a clear definition of what "slow down" even means. Without technical specifications or measurable milestones, it is hard to tell a statement of intent from a PR exercise.

Ha pointed out that even before Amodei's post, many in the industry had already adopted similar versions of the idea. Those proposals have been percolating in the AI safety community for some time, but the speed with which consensus coalesced around them raises eyebrows. On one hand, broad agreement is encouraging; on the other, it suggests the steps will not be particularly dramatic, or they would not have attracted such sweeping support.

Korosec warned that such consensus could be the start of a cartel. She stressed the discussion concerns frontier labs only — a narrow and important distinction. The general idea involves independent external evaluators embedded inside companies like Anthropic or OpenAI to monitor safety practices and incidents, yet the mechanics of that arrangement are vague. Another proposal envisions safety coordination among large AI companies in democratic countries, again with no practical detail on how it would work.

O'Kane questioned whether existing regulation or market forces can offer adequate protection. The current federal administration, he said, shows little appetite for enforcement in general, let alone in this specific domain. The consumer-choice mechanism fails, too: there is not enough genuine competition to make a company pay a price for irresponsible behavior. Mass cancellations or user abandonment simply do not happen when a handful of dominant players control the market.