Altman says AI can be slowed as industry splits over safety and regulation

OpenAI chief executive Sam Altman used his appearance at the Dreamforce conference in San Francisco to argue that artificial intelligence can be slowed before it becomes uncontrollable. He acknowledged that accidents will happen as the technology develops, but said he is confident in his company's ability to slow down AI before it gets out of control.

Altman said the industry should coordinate so it has enough time to develop the technology safely. He also acknowledged public fear that AI companies could become too powerful. He said, 'The world should trust that we are going to do the right thing because it's the right thing and because we feel the magnitude of this.' He warned that fears grow when commercial pressures or a race between companies or countries creates the impression that some actors will not act responsibly.

Other industry leaders gave different answers. Nvidia chief executive Jensen Huang said tech companies should judge the safety of their own products. If a company is not confident about a product's safety or capability, it should not release it, he said. Huang argued that market forces already provide discipline and that no new laws or regulations are needed. He added that innovation, speed and safe products are not a false choice.

Meta chief executive Mark Zuckerberg posted on X that every lab has the responsibility and incentive to train its models safely and to act on its own to ensure that happens. He said trust and alignment are becoming key capabilities that will differentiate AI agents.

The debate has also moved into politics. President Trump has continued to downplay concerns about AI and rejected calls to slow its development, saying whoever wins AI wins. At a rally in Washington, progressive Senator Bernie Sanders and former Trump adviser Steve Bannon called for restrictions. Sanders warned that the potential dangers of AI are probably greater than nuclear weapons. Bannon accused companies of trying to socialise risk and keep the upside.

AI expert Gary Marcus, author of Taming Silicon Valley, was sceptical of much of the discussion. He said Altman and Huang have a lot of money at stake and that asking their companies to pull back voluntarily is like asking cigarette companies to stop selling cigarettes. He criticised OpenAI for releasing a model, GPT-6 Astra, that he said is less monitorable, after an incident involving Hugging Face that could have been prevented with better monitoring. He also said Sanders was exaggerating the comparison with nuclear weapons, though he sees serious cybersecurity risks. Marcus repeated his call for an independent regulator similar to the FDA, with independent scientists deciding whether the benefits of an AI system outweigh its risks. He said the White House has a secret vetting process and that almost all of 132 pages released under a Freedom of Information Act request were redacted. The Financial Times also reported that OpenAI is seeking another private funding round to raise tens of billions of dollars.