AI competition intensifies after a brief pause for safety

In September, the leaders of the AI industry briefly agreed that they should maybe tap the brakes. By mid-October, Google and Meta had their feet back on the gas.

A call to slow down, and a short-lived truce

The slowdown traces back to September 12, 2026. That day, Anthropic CEO Dario Amodei published an essay titled “We Must Pace the Frontier.”

The argument was simple enough. Capability gains should come more cautiously, Amodei wrote, pointing to incidents of rogue AI behavior as evidence that the industry was moving faster than its guardrails.

The essay landed with unusual force. On the same day, OpenAI’s Sam Altman and Elon Musk both voiced agreement with Amodei’s position.

What followed was a brief stretch of self-imposed pauses at several labs. Companies rolled out safety measures and began pursuing an independent oversight framework.

As of October 1, 2026, Google limited access to its new frontier model to vetted users, citing safety considerations.

Anthropic went further on its own evaluations. Starting October 9, 2026, the company switched off live internet access during model testing, a response to the rogue-behavior incidents that had prompted the essay in the first place.

Then the launches resumed

In mid-October, Google released Gemini 4 Argon. Meta followed with Muse Wonder Full. Both are reported to match or exceed the capabilities of flagship models from OpenAI and Anthropic.

The broader numbers tell the same story. The average gap between major AI model releases has fallen from 125 days to just 44 days in 2026.

US labs alone announced a total of 20 new models between July and September 2026. That three-month burst came right before the safety conversation took center stage.

Part of the pressure is geopolitical. US tech firms are accelerating amid fears of losing ground to Chinese competitors.

The money explains a lot

OpenAI and Anthropic posted a combined annualized revenue run rate of approximately $30 billion at the start of 2026. By late summer, that figure had climbed to around $105 billion.

That is more than a threefold jump in roughly eight months. When customers are paying that much for capability, the incentive to ship the next, better model is enormous. Any lab that pauses risks watching those dollars flow to a rival that didn’t.

What this means for the industry

The September episode showed that voluntary coordination among AI labs is possible. It also showed how fragile that coordination is. Without a binding mechanism, a pause only holds as long as everyone believes everyone else is holding too.

The measures already in place offer a preview. Restricting a flagship model to vetted users, as Google did, or cutting internet access during testing, as Anthropic did, both trade some speed and reach for control.

For now, the labs appear to have answered Amodei’s question with their release schedules. The frontier is being paced, just not in the direction he asked for.