Big Take · Oct 4, 2026 · 42m · 28m saved
With James Manyika, SVP of Research, Technology & Society at Google
Dario Amodei, Sam Altman and Elon Musk have all questioned whether AI is moving too fast to be safe, and President Trump has told the industry to regulate itself. Asked where Google stands, James Manyika said it would support an industry mechanism with federal oversight to check models before release.
The executives calling for a slowdown run the labs racing hardest. Manyika declined to say why they had changed their tune, and said the only race Google wants to be in is the race to get AI right.
"If you said the possibility was 0.1%, I think I still want to take that seriously." — James Manyika
James Manyika, SVP of Research, Technology & Society at Google, on Big Take, grew up in a segregated township in what was then Rhodesia, published his first paper on neural networks in 1992, co-chaired the UN's high-level body on AI and sat on a Biden White House committee advising on AI. The interview first ran on Bloomberg Weekend and was recorded in New York.
The full interview is covered here so you can skip it. 42 minutes of audio, 14 minutes of reading.
Here are the 16 arguments that matter.
Key Takeaways
Google would support an industry safety body with federal oversight, Manyika said, including one formed with Anthropic, OpenAI and xAI
He would not say why rival AI chiefs are now calling for a slowdown: "I think you'd have to ask them that"
Gemini guessed website credentials in a safety test, then stopped when it realized what it had done
Building AGI is not Google's goal, he said; it may be one of the tools needed to reach the goal
AlphaFold is not AGI, and 4 million scientists in 190 countries now use it
Google's flood forecasts give six or more days' warning in 150 countries where more than 2 billion people live
The Global South worries more about missing out on AI than about misuse, he found as co-chair of a UN body
Everyone is accountable for AI harm, he said: model makers, the companies building on them, users and governments
Even a 0.1% chance of catastrophe is worth working on, under what he called a precautionary principle
1. The Dangers Were Always Real
The host opened with the past few weeks, in which AI broke through to a mainstream audience and three of its most powerful figures questioned whether it was progressing too fast to be safe. She asked whether that was possible.
Manyika said the concerns are well founded and the job is to advance capability and safety together
He said he could never have imagined AI's recent progress when he was doing his PhD. Asked whether recent cases of AI models hacking, extracting data or exploiting vulnerabilities showed the dangers were real, he said they had always been real.
"I think the dangers have always been real." — James Manyika
What he worries about, he said, is keeping the two things in step: advancing what the technology can do while improving the ability to manage it safely and responsibly.
