‘We can do this safely’: Sam Altman backs AI industry’s ability to manage risks

OpenAI chief executive Sam Altman has expressed confidence that artificial intelligence companies can continue developing increasingly capable systems while managing safety risks, as debate intensifies over whether governments should impose stronger controls on the technology.

Speaking with Salesforce chief executive Marc Benioff at the Dreamforce conference on September 15, Altman said he was confident in both OpenAI and the wider AI industry's ability to develop the technology safely without causing significant harm to people. 

“I am very confident in our company’s ability – our industry’s ability – to do this safely,” Altman said.

His comments come amid renewed discussion among AI leaders, policymakers and researchers over whether the pace of development is moving faster than the industry's ability to manage emerging risks.

The latest debate was prompted in part by Anthropic chief executive Dario Amodei, who published an essay recently warning about the potential risks of rapidly advancing AI and calling for a slower pace of frontier AI development. 

Amodei has proposed a three-step approach involving stronger safety practices within AI companies, industry-wide standards and international cooperation. He reiterated the approach at Dreamforce, calling for greater transparency, investment in safety and common standards across the industry. 

Altman said OpenAI is already working with Anthropic and Google's DeepMind on AI safety issues, but argued that AI companies should be capable of improving their models safely without relying solely on government intervention.

“I’m disappointed by how it’s been framed,” Altman said, referring to the debate around the pace and safety of AI development. 

The discussion comes as AI systems become increasingly capable of performing tasks with greater autonomy. Altman has described the technology as moving towards a stage where AI systems could carry out more work independently, while acknowledging that widespread adoption will also depend on organisations changing established workflows. 

The debate has exposed differing views among technology leaders on how AI safety should be managed.