At Dreamforce in San Francisco, Nvidia chief executive Jensen Huang argued that AI companies need no laws or regulations to slow development, while Anthropic chief executive Dario Amodei made the case for outside evaluation and coordination. Their opposing proposals put a clear choice on the table: trust companies to pause products they consider unsafe, or build shared oversight before the technology moves faster still.
Jensen Huang Watch analysis
What happened
The exchange took place during Salesforce’s Dreamforce event, held from 15 to 17 September, according to BigGo Finance’s report, published on 13 October. Huang said safety is an engineering problem and argued that companies can test systems and hold back products they do not consider safe without new laws.
Amodei set out a three-stage approach: independent evaluators with access to AI companies, coordination among democratic countries on safety standards and development pace, then international coordination. He said Anthropic had committed to the first stage and planned to engage with other industry players.
Why it matters
These are not just two different views of regulation. Huang’s approach puts the key judgement with the companies building the systems: they test, decide whether a product is safe, and pause if needed. Amodei’s proposal adds outside scrutiny and coordination, on the premise that firms may need common rules when their incentives and risks overlap.
That distinction matters as AI systems become more capable and competitive pressure grows. A company’s internal testing may be fast and technically informed; independent access could make it harder for important risks to remain inside the building. Neither approach, on its own, settles how oversight should work in practice.
Our read
Huang offers a tidy engineering answer to a question that is also about accountability: who gets to decide when the testing is enough? Amodei’s staged plan gives that question a more concrete shape, starting with external evaluators rather than leaping straight to a global rulebook. The hard part is making scrutiny meaningful without turning it into a slogan or a bottleneck.
What to watch
- Whether Anthropic gives more detail about how its independent evaluators will work.
- Whether other AI companies agree to comparable outside access.
- Whether governments propose standards that reflect either leader’s approach.
Discussion spark: Should decisions about whether frontier AI is safe enough to release rest with the companies building it, or should independent evaluators have a formal role?
Sources and evidence
Independent WittyWires tracker for public updates about Jensen Huang. Not affiliated with or endorsed by Jensen Huang; this is not an official account.