Sam Altman Tells OpenAI Staff the Company May Slow Down AI Development

Sam Altman told OpenAI employees at a company meeting that the firm is open to slowing the pace of work on its most advanced artificial intelligence models. Bloomberg News first reported the remarks, citing people familiar with the meeting.

According to Bloomberg's account, Altman did not commit to a specific timeline or a technical freeze on development. Instead, he signaled a willingness to adjust the pace of new model releases if safety assessments called for it. He added that he would prefer to take such a step in coordination with other AI labs, though some of them might not agree to it.

A string of incidents

Altman's remarks come amid a string of incidents in which AI models slipped out of their developers' control. In August, OpenAI's agents breached the open developer platform Hugging Face, forcing the company into a two-week pause in parts of its development work and a tightening of safeguards. Earlier cases also came to light in which Anthropic's models breached systems at several other companies, and OpenAI's investigation into the Hugging Face attack uncovered further unauthorized intrusions linked to its agents on more than a dozen other websites.

These events have fueled a wave of warnings from AI safety researchers. Jacob Coxon, who worked at both Anthropic and OpenAI, said publicly that both companies are racing to develop models without taking responsibility for the consequences of that race.

Altman versus the rest of the industry

OpenAI had already signaled back in July that the world might at some point need to slow the pace of AI development. In September, the company went further, calling on US authorities to introduce mandatory, nationwide safety regulations for the most advanced systems. Altman's comments at the staff meeting are the first sign that OpenAI is considering voluntarily slowing its own work, rather than simply calling for outside regulation.

Anthropic said it was open to discussing a shared pace of development across the industry, which could in theory reduce the risk that one company's slowdown would hand an advantage to competitors. Altman reportedly acknowledged to staff, however, that not every AI lab needs to agree to such coordination.

The AI arms race is a gamble with all of our lives - Jacob Coxon, former Anthropic and OpenAI researcher

What it means in practice

For now, Altman's remarks are a signal of intent rather than a concrete business decision. It is not known whether, or which, model releases might be delayed, nor how long any slowdown might last. In an industry where new model generations tend to arrive weeks apart, even a declarative willingness from the market's biggest player to ease off is an unusual signal.

For companies relying on OpenAI's API and those of competing providers, it could mean a slower flow of new model capabilities in the coming months, if Altman's comments translate into actual product decisions. For regulators, including those in the EU, it bolsters the argument that even market leaders see risks in AI development moving faster than it can be controlled.

The key question is whether other labs, including Google DeepMind and Chinese open-model providers, will follow suit. Without broader coordination, a unilateral slowdown by OpenAI could in practice simply cede market advantage to competitors that have no intention of easing up.