What the AI labs said next: pacing, not stopping

In September 2026, Anthropic CEO Dario Amodei published an essay responding to the incidents. His central ask:

“We must slow the pace at which we improve the capabilities of AI models.”

He was explicit about what that does not mean:

“pacing does not mean halting model training or technical progress”

And about how much the builders understand:

“we still only understand a tiny fraction of what goes on inside these models”

The forecast, labeled

The line most often repeated from the essay is a projection. Here it is in full, so it is not shortened into something he did not say:

“it's my worry that in 6-12 months such a swarm could be capable of taking over the entire internet with a persistent botnet (potentially causing hundreds of billions of dollars in damage)”

A botnet is a network of hijacked computers under one controller. This is his worry about the next 6 to 12 months. It is not an event, and we label it as a forecast everywhere on this site.

OpenAI slowed down too

OpenAI paused some training after the Hugging Face incident. In its own words, the slowdown included

“a two-week pause in reinforcement learning (RL) training on our latest models intended for deployment”

and it said its largest planned frontier training run remained on hold while it gathered more evidence of alignment. OpenAI also called the incident

“a “warning shot” for us and for the world”

What a lab did about its own record

Anthropic also reviewed its own evaluation history after OpenAI's disclosure and published what it found, including cases where its models attacked real systems through misconfigured tests. See the incident page.

Context

Amodei's earlier writing makes the case for the upside of AI as strongly as the risks. Reading both is the fairest way to understand why a lab leader would ask to slow down without asking to stop.