Sam Altman urges caution on AI; Trump says US must keep its lead over China

OpenAI CEO Sam Altman said every frontier AI lab has a responsibility to demonstrate that it can develop increasingly capable systems safely

The rapid progress in artificial intelligence (AI) must be accompanied by stronger safety measures, OpenAI founder and CEO Sam Altman said on Monday, arguing that AI capabilities should not outpace alignment and monitoring.

 

“The world deserves confidence that American companies developing increasingly capable AI will act responsibly, especially as the trajectory of progress has steepened,” Altman said in a post on X.

 

He said every frontier AI lab has a responsibility to demonstrate that it can develop increasingly capable systems safely. While Altman welcomed a federal framework establishing consistent safety requirements for frontier AI, he said companies should not wait for legislation to begin strengthening safeguards.

 

 

Existing safety frameworks, including Responsible Scaling Policies and Preparedness Frameworks, were largely focused on the deployment of completed models, Altman said. The increasing focus on safe development and evaluation, he added, requires new tools.

 

OpenAI now formulates explicit safety cases before frontier reinforcement learning runs that are expected to significantly increase a model’s capabilities. Altman said these measures are in addition to the safety work conducted before model releases.

 

The company also called for shared industry standards covering misalignment, monitoring and safety, saying these could lead to better outcomes. OpenAI said it hoped other AI companies would develop and share their own approaches.

 

What does Altman mean by ‘pacing’ AI?

 

Altman clarified that “pacing” AI development does not mean stopping technological progress.

 

“Progress has been rapid and will continue to be. But it should be slower than it otherwise could be,” he said, noting that measures such as safety cases and monitoring carry significant costs.

 

Altman said slowing the pace would be worth that cost, arguing that “no amount of American competitive pressure should justify recklessness, or let capabilities get ahead of alignment and monitoring.”

 

He said government assistance would be particularly important for international coordination, but added that AI developers should first take the steps they can implement themselves.

 

Altman’s remarks came after he, Elon Musk and Microsoft CEO Satya Nadella backed Anthropic CEO Dario Amodei’s call for a global slowdown in AI development. Altman had earlier agreed with Amodei that the industry needs to “pace the frontier”, saying the issue had been a primary topic of discussions at OpenAI in recent weeks. 

 

Trump takes opposite stance

 

US President Donald Trump, however, took a different view, stressing the need for the US to maintain its lead over China in AI while acknowledging that some guardrails could be introduced.

 

Addressing reporters during his trip to the Doonbeg golf course in Ireland, Trump said the US could put “guardrails” on AI but emphasised the importance of retaining its international leadership in the technology.

 

“Look, we’re leading China in AI … and, frankly, I want to keep it that way, because whoever wins AI wins,” Trump said, quoted by AFP.

 

Trump added that while the US could introduce guardrails, he was concerned about “negative forces” pushing AI in directions that may not be realistic or desirable.

 

The debate over the pace and safety of AI development has intensified in recent weeks, following warnings from researchers and industry leaders about the risks posed by increasingly capable AI systems. Anthropic researcher Jacob Coxon, who quit the firm last week, said that companies are racing to build AI systems they “may not be able to control.” 

 

The concerns have also been fuelled by incidents involving AI agents. In July, OpenAI models circumvented controls during internal cybersecurity evaluations and gained access to external systems, including Hugging Face. OpenAI later said the incident showed that highly capable AI agents can work around technical controls, collaborate through unauthorised channels and take actions that were not directed by humans.