AI leaders push for safety and transparency, split on need for new rules

The heads of three of the world's most prominent artificial intelligence companies used a single stage in San Francisco to argue that safety and transparency must sit at the centre of how the technology is built, even as they differed on whether governments need to step in.

Nvidia chief executive Jensen Huang told an event hosted by Salesforce chairman and chief executive Marc Benioff that safety is "job one" and should be treated as an engineering challenge rather than an abstract principle. Because AI systems are ultimately computing systems, he said, the priority is to build sound test environments in which complex models can be evaluated before release.

Huang said companies should hold back a product if they are not confident about its functionality, capability or safety, and that the pace of development should be set accordingly. Speed and safety, he argued, are not a choice between two options — both can be pursued at once. Where a product or a company appears to be getting out of control, he added, the right response is to pause and fix the problem.

He also pushed back on the idea that fresh legislation is needed, saying market forces already create sufficient pressure on firms to release only products they trust. In his view, companies simply need to make that decision for themselves.

OpenAI chief executive Sam Altman struck a different note on the question of conditionality. He said safety commitments should carry no qualifiers, and that the public should be able to trust firms to act responsibly regardless of what competitors or other countries do. Altman described it as a positive development that the industry wants to coordinate and give itself enough time to build the technology safely, but warned that any suggestion that commercial or geopolitical competition might lead some players to cut corners is what alarms people most.

Anthropic chief executive Dario Amodei, whose company helped set off the wider debate over the pace of development, called for stronger internal practices and greater openness. Drawing an analogy with the car industry, he said the more responsible response to a rival's safety failure is not to attack the competitor but to examine one's own record, improve practices, recommit to transparency and invest further in safety.

The discussion reflects how a question once confined to principle — first raised publicly when Anthropic researcher Jacob Coxon resigned while warning that building superintelligent systems posed a genuine risk to humanity — has grown into a broader argument over the direction the AI industry will take.