Nvidia CEO Jensen Huang said the chipmaker "absolutely must" work with Chinese AI companies on monitoring risks around what he called a "potentially dangerous" technology. He made the remarks to CNBC at a media briefing in London on Wednesday evening, October 14, according to CNBC's report published October 15, 2026.
"We absolutely should work with each other so that we advance the technology safely," Huang said. The comment is a statement of position from the head of the company that supplies much of the world's AI compute, not an announced program. This post lays out what was said, the policy backdrop, and the questions the remarks leave open.
TL;DR
| Question | Answer (source: CNBC) |
|---|---|
| What was said? | Nvidia "absolutely must" work with Chinese AI companies on monitoring risks |
| Where and when? | CNBC media briefing in London, Wednesday evening, October 14, 2026 |
| Mechanism | "Dialogue"; Huang says he has "dialogues with many of the researchers in China" |
| Who else should do it? | "The United States should do it... I advocated that the U.K. do the same thing" |
| Any program announced? | No program, partner, date or budget in the report |
| Huang's regulatory stance | Described by CNBC as a prominent opponent of more AI regulation |
The quotes in context
CNBC's key points are short. Huang called AI "potentially dangerous," said he has ongoing "dialogs" with researchers in China, and said: "We ought to enhance the dialogue. The United States should do it... I advocated that the U.K. do the same thing." He added that the UK is positioned "in the world to be able to be a convener of many of these dialogues" because of its "intellectual capital."
Two things are notable. First, "work with" is defined as dialogue and monitoring risks, not joint training, shared evaluations or hardware sales. Second, the audience was a media briefing, not a policy forum, so the remarks are a headline position rather than a proposal with terms.
The backdrop: governments are already talking
CNBC places the remarks against rising attention to AI safety. It says researchers from AI labs including OpenAI and Anthropic have in recent months called for a slowdown in developing the technology and warned of existential risks, and that OpenAI's Sam Altman and Anthropic's Dario Amodei have called for greater regulation, while Huang "has emerged as a prominent voice in opposition to that view."
On the government side, CNBC reports that before President Donald Trump's meeting with China's Xi Jinping, Treasury Secretary Scott Bessent discussed the prospect of a "U.S.-China AI dialogue" with his Chinese counterpart He Lifeng in September, including a channel for incidents "up to a national security level." A Chinese readout of the meeting quoted Xi as saying the countries "may have competition, but they may cooperate even more" in AI, and that the two sides "can continue their dialogue on AI, exchange views on its risks and benefits, and jointly prevent the misuse and abuse of AI."
Huang has made similar arguments before: the South China Morning Post reported on an earlier appearance in which he said competitors are still able to collaborate, at a time when the Trump administration had reportedly rejected a joint US-China AI safety framework. So Huang's call echoes a conversation that already exists between governments. What he adds is the voice of a company whose products are central to export-control debates.
Why Nvidia's position is distinctive
Nvidia sits in an awkward spot between the two countries. It sells the accelerators that frontier training depends on, its sales into China are shaped by US export rules and Beijing's own restrictions, and its CEO has repeatedly argued that competition and cooperation can coexist. His statement fits a pattern. Yesterday we covered his comments to Bloomberg that he would stop supplying chips to a customer building unsafe AI, and our analysis of why that statement is a hypothetical rather than a policy is in the claimed-versus-verified post. Earlier remarks include his claim that AGI has arrived.
The pieces together describe a consistent line: AI is powerful and potentially dangerous; dialogue and cooperation are the right tools; heavier regulation is not. Whether you agree is a policy judgement, and critics from the safety community argue that voluntary dialogue without verification or enforcement is insufficient. The report itself does not quote critics, and we do not attribute criticisms to named people here.
The UK angle: a safety pledge, and the "convener" idea
Huang said the UK could convene such dialogues. That lines up with Nvidia's commitments announced during his European tour: a five-year, $250 million compute pledge for UK agentic AI safety research, which we examined in NVIDIA's $250M UK safety compute post, and Europe-wide commitments covered in Nvidia's AI-in-science pledge. It also echoes the UK's earlier role in safety evaluation, including the White House's restoring of UK AI Security Institute access to frontier models.
Placing a US-China channel in London has a logic: the UK has a safety institute, hosted the first major AI safety summit, and is not a party to the chip-export dispute. But nothing in CNBC's report says the UK government has agreed to take such a role, and no Chinese company or researcher was quoted as responding.
What "dialogue" could mean in practice
Huang did not define the word, so it helps to separate the forms cooperation could take. The list below is our illustration of the options, not a description of anything Nvidia or either government has proposed.
| Form | What it would involve | Hardest part |
|---|---|---|
| Researcher exchange | Scientists from both countries meet, publish and compare risk assessments | Visa, export and security restrictions; sensitivity of frontier work |
| Incident channel | A line for reporting serious AI incidents between governments, as Bessent reportedly discussed for incidents "up to a national security level" | Agreeing what counts as an incident and who is told |
| Shared evaluations | Common tests for dangerous capabilities such as cyber or bio misuse | Trust, test-set secrecy, and who runs them |
| Red lines | Agreed prohibitions, for example on certain autonomous weapons uses | Verification and enforcement |
| Industry safety forums | Companies from both sides share practices | Antitrust, sanctions and political risk for the firms involved |
Huang's own words point to the first row. He spoke of "dialogues with many of the researchers in China" and of enhancing dialogue, which is the lowest-commitment form. Nothing in CNBC's report suggests the others.
Why this matters now: the weeks around the remark
Two threads converge. The first is the safety debate inside the labs, which CNBC summarises as researchers calling for a slowdown and warning of existential risks. We have covered the other side of that debate too, including a Google DeepMind researcher's claim that self-improving AI is not close and the EU AI Office's formal probes into OpenAI and Anthropic after agent incidents. The second is the geopolitical thread, in which chips, export rules and model access are bargaining chips between Washington and Beijing.
Huang's remark sits at the intersection. When the head of the dominant chip supplier says cooperation with Chinese firms on safety is a must, it signals that the industry wants safety framed as a shared problem rather than as a reason for tighter control of exports. That is a defensible position and also a self-interested one, and readers should hold both in mind. A company that benefits from selling compute widely has an obvious interest in policy that does not restrict sales; that does not make the safety argument wrong, but it explains why Huang and the lab leaders who favour regulation land in different places.
The disagreement with other lab leaders
CNBC sets Huang against Altman and Amodei, who have called for greater regulation. The disagreement is not whether AI is risky; Huang himself calls it potentially dangerous. It is about the remedy: dialogue and voluntary cooperation versus binding rules. For builders, the practical consequence is that the regulatory environment is still unsettled, and plans should assume both outcomes: voluntary norms in some places, binding rules in others, and incident-reporting duties that may arrive faster than capability limits.
What is claimed, and what is verified
| Item | Status |
|---|---|
| Huang said Nvidia "absolutely must" work with Chinese AI companies on monitoring risks | Confirmed by CNBC's report, which quotes him directly |
| He has ongoing dialogues with researchers in China | His own statement; not independently verified |
| A formal Nvidia program with Chinese companies | No evidence; not announced |
| Bessent and He Lifeng discussed a US-China AI dialogue | Reported by CNBC for September; details of the incident channel come from that report |
| Xi's remarks on AI cooperation | From a Chinese readout, as relayed by CNBC |
| Share-price reaction | One aggregator claims a premarket drop of 1.8% tied to the comments; we could not verify that causal link, and market moves have many causes |
What this means for builders and policy watchers
For builders, nothing changes today: no new export rule, no new standard, no shared evaluation. The practical reading is directional. Industry leaders who oppose binding rules are nonetheless saying they want to see safety dialogues, which suggests the next phase of AI governance will be negotiated through channels like the UK-convened talks and US-China incident hotlines rather than through chip-level controls.
If you build or deploy agents, safety also means concrete controls at runtime, not just diplomacy. AgentBeam is the agent security platform from the explainx.ai team that stops AI agents before they take dangerous actions. For the policy side, follow how any US-China channel treats incident reporting, since an incident channel is the part of Bessent's discussion that could eventually touch deployed systems.
What to watch next
- Whether any Chinese lab, ministry or researcher responds publicly to Huang's call.
- Whether the UK government confirms a convening role or hosts talks.
- Whether Washington's position on a US-China AI dialogue hardens or softens after the Trump-Xi meeting.
- Whether Nvidia attaches a concrete mechanism, such as shared evaluations or incident sharing, to the "work with" language.
- Any change to export rules that would constrain what cooperation is legally possible.
Related reading
- Jensen Huang would halt chips to an unsafe AI customer: claimed vs verified
- Jensen Huang: AGI has arrived
- NVIDIA's $250M UK agentic safety compute pledge
- Nvidia's Europe AI-in-science pledge
- UK AI Security Institute regains frontier model access
- EU AI Office probes OpenAI and Anthropic agent hacks
Details are accurate as of October 15, 2026, from CNBC's report; no program or timeline was announced.