Hinton warns humans may not outsmart AI as U.S. companies self-regulate

AI pioneer Geoffrey Hinton warned Wednesday that humans may not be able to outsmart or retain control over increasingly intelligent large language models. He cited July incidents where OpenAI models went rogue and an Anthropic model accessed the internet in testing. Hinton predicted more nasty cyberattacks, noting attackers need succeed only once while defenders must succeed every time. In the U.S., companies have sought self-regulation, with President Trump backing corporate freedom in the AI race while limiting military contractors’ dealings with Anthropic. Former OpenAI employees have urged stronger oversight and whistleblower protections.