Huang Renxun and the heads of Anthropic and OpenAI argue again about AI security: market power is sufficient to protect, no new regulation is required

The Zhitong Finance App learned that on the stage of the Dreamforce Conference in San Francisco, the dispute over artificial intelligence (AI) safety and regulation routes is moving from behind the scenes to the front of the stage. On Tuesday EST, NVDA.US (NVDA.US) CEO Hwang In-hoon, Anthropic CEO Dario Amorday, and OpenAI CEO Sam Altman once again expressed very different views on AI security at the CRM.US (CRM.US) annual Dreamforce conference. This disagreement occurred just a few days after the AI security debate entered the mainstream view.

Amouday was the first to take the stage with SEFTSE CEO Mark Benioff. Just last weekend, he published a 3,800-word article urging the industry to slow down model development, and reiterated that position in his keynote address on Tuesday.

“I think this is the way to lead the industry forward — set an example and tell everyone that we can always do better,” Amodei said to an audience of about 12,000 people at the Moscone Center in San Francisco.

Amodei proposed a three-step plan aimed at slowing the rate of improvement in model capabilities without “sacrificing commercial advantage or America's leading position in the AI field.” His proposal was quickly supported by a number of industry executives, including Altman, SpaceX (SPCX.US) CEO Elon Musk, and Google (GOOGL.US) DeepMind Chairman Demis Hassabis.

The reason behind this statement is that last week some industry researchers warned that AI could cause catastrophic harm. Earlier, an Anthropic employee made a high-profile resignation, saying that the company and rival OpenAI were “risking our lives” because their technology could pose a threat to humans.

Shortly after Amodei took the stage, Hwang In-hoon also appeared on the same stage as Benioff, but he brought a very different perspective.

The CEO of the world's highest company by market capitalization made it clear that market forces already exist and that the industry does not need new laws or regulations. “We don't need new laws — we don't need new regulations,” Wong In-hoon said bluntly at SAIFS's event.

He further refuted the view that AI development must choose between safety and speed: “This is a false proposition. You can have both at once.”

Hwang In-hoon believes that acting responsibly depends on the model developers themselves. “Run as hard as you can,” he said, “but if at any moment you feel the company is out of control or the product isn't secure enough, then stop and make sure you're doing things right.”

“Safety is paramount,” he emphasized. In many ways, this is the number one priority. However, safety is an engineering issue.”

A day ago, Hwang In-hoon received a call from US President Trump on the stage of the All-In Summit in Los Angeles. During the call, Trump expressed concern about those trying to hinder the construction of AI data centers. “We won't let this happen, sir,” Hwang In-hoon said to Trump.

Also, Altman had a fireside conversation with Benioff later that day. During the conversation, he said that given the strong capabilities of the current model, safety and monitoring must be prioritized over capacity improvement.

“Some companies will say, 'We are only responsible if other companies are also responsible',” Ultraman said. People will then ask what this company will do “no matter what” to ensure safety and responsibility.

“There should be no strings attached to this,” Ultraman stressed.

As host, Benioff said he respects each company's own judgment: “If they think they should slow down, then slow down. If they feel they should speed it up, then speed it up. Then they should be held to account.”

Business cooperation progresses simultaneously

It is worth noting that, in the midst of intense debate, commercial cooperation is also progressing at the same time. On Tuesday, Seflex and Nvidia announced the launch of a new inference model for Saiflex's Agentforce platform, which is built on Nvidia's open source Nemotron model. At the end of last month, Seflex and Anthropic expanded their strategic partnership and launched “Claudeforce” — a new way for sales people to access critical data directly from within Anthropic's Claude chatbot.

The Dreamforce Conference has been the software industry's top event for over a decade, attracting around 50,000 people to downtown San Francisco every year. Benioff's keynote speeches have always been famous for inviting a number of high-profile guests, who are often central figures in today's tech world. Benioff has long-standing cooperative relationships with both Hwang In-hoon and Amoudai.

The debate reflects deep divisions in the AI industry over security and regulatory issues. One side, represented by Amodei and Ultraman, argues that in the context of rapid improvement in model capabilities, the industry should take the initiative to slow down, strengthen safety monitoring, and even sacrifice some commercial interests. The other side, represented by Hwang In-hoon, firmly believes that market power is enough to drive companies to innovate responsibly. Excessive regulation may stifle America's leading edge in the AI field.

As AI technology becomes more deeply entrenched at all levels of society and economy, the “speed and safety” debate is expected to continue to heat up and may have a profound impact on future policy trends.