Dario Amodei outlines three-step approach: internal review, industry standards, international cooperation

'AI is advancing far faster than expected — but slowing down doesn't mean stopping'

Anthropic CEO Dario Amodei speaks on stage during the keynote address by Salesforce CEO Marc Benioff at Dreamforce, Salesforce's annual conference, at the Moscone Center in San Francisco on Tuesday (local time). [AFP]
Anthropic CEO Dario Amodei speaks on stage during the keynote address by Salesforce CEO Marc Benioff at Dreamforce, Salesforce's annual conference, at the Moscone Center in San Francisco on Tuesday (local time). [AFP]

Dario Amodei, the CEO of Anthropic who has called for a more measured pace of AI development, said AI safety should be approached the same way automakers manage vehicle safety.

Speaking at Dreamforce, Salesforce's annual conference at the Moscone Center in San Francisco on Tuesday (local time), Amodei took the stage alongside Salesforce CEO Marc Benioff and outlined a three-step framework for AI safety: internal review, the establishment of industry-wide standards, and international cooperation.

"Imagine a car company discovers a safety problem — a brake issue or a manufacturing defect," he said. "The first responsible step is to look at your own record."

Companies should first audit their own practices, improve existing procedures, reaffirm transparency and increase investment in safety, he said.

"After that, the whole industry needs to come together and discuss safety standards that apply to everyone," he said. "And finally, there needs to be international cooperation."

Amodei said the safety plan he has proposed for AI follows exactly that model.

He noted that while he had previously predicted AI would grow more capable as computing resources expanded — the so-called scaling law — even he had not anticipated how fast that progress would come.

"I'm surprised that this kind of speed is possible in the business world," he said.

He added that even if AI development were frozen at its current state, the value society is actually capturing represents only 5 to 10 percent of what AI is capable of delivering. "There are still endless areas where it can be deployed more broadly to provide social benefits," he said.

Amodei was careful to clarify that calls for a slower pace do not mean halting progress. "Talking about slowing down AI doesn't mean freezing the technology," he said. "AI will continue to advance."

Benioff echoed the sentiment. "What has happened over the past 10 years has been truly remarkable — no one could have predicted this scenario," he said. "We need to keep a close eye on the pace of AI development."

Amodei sparked widespread debate Saturday when he published a post on his personal blog arguing that AI development should be slowed in the interest of safety.

OpenAI CEO Sam Altman, Google Chief Scientist Demis Hassabis and SpaceX AI CEO Elon Musk have expressed support for the position, but President Donald Trump has dismissed such AI safety concerns as a "scam" and strongly opposed calls for AI regulation.


mokiya@heraldcorp.com