Silicon Valley Divided on AI Development: Pause vs. Accelerate

Original author: Su Yang

Original editor: Xu Qingyang

Source: Tencent Tech

Cutting-edge AI continues to accelerate or hit the brakes, becoming the most contentious topic in Silicon Valley today.

Amid the controversy, Silicon Valley giants have taken sides: Altman, Otterman, Musk, and Hassabis form the “Pause Camp.” Huang Renxun is a staunch member of the “Accelerate Camp,” criticizing AI doomsday theories as lacking scientific basis; Yann LeCun has consistently mocked Altman, saying: “When he was still at OpenAI, he argued that GPT-2 was too dangerous to open-source—I laughed at him then, and now everyone can continue.”

In contrast, Zuckerberg, Nadella, and others have remained relatively neutral.

The giants have taken different sides, indicating that Silicon Valley has not reached a consensus on pausing frontier models, stemming from differences in their ecological positions within the AI industry chain. While leaders at the model level may call for a pause for various reasons, it is unlikely that companies like Oracle and AWS, which hold hundreds of billions of dollars in computing orders, or NVIDIA, at the top of the value chain, will respond positively to such a request.

In Washington, the call for a pause not only failed to gain support but also triggered a series of questions.

“It feels a little bit to me like a bit of a Trojan horse,” JD Vance said in an interview, a sentiment that reflects the real view in Washington: when companies building the Trojan horse repeatedly ask for the door to be made narrower, it’s time to examine whose interests this door truly serves.

The spark for this debate was the resignation of Anthropic researcher Jacob Cockson.

He spent three years conducting pretraining research at OpenAI and Anthropic, directly involved in training the most advanced models. In a long post on X, he offered no pleasantries: “Neither company has acted responsibly. They are racing headlong toward self-improving superintelligence, betting our lives on it.”

He bluntly stated that people inside the lab genuinely believe AI could kill all humans within a decade, and this is far from a public relations stunt. Many colleagues have already begun describing the current stage as a “tipping point” or “endgame.” Coxen even gave up his upcoming equity stake; he had only been at Anthropic for four months, and company policy requires six months to vest.

Jacob Coxon, former researcher at Anthropic

Immediately following, Evan Hubinger, responsible for alignment science at Anthropic, publicly responded: Jacob is right—we genuinely believe AI could kill all humans. I personally think the probability of this happening within ten years exceeds 10%. Anthropic is trying its best, but we still lack a clear plan for aligning superintelligence and are far from being on the right track.

These two gunshots brought concerns that had previously been confined to internal laboratory discussions directly into the public eye. Against this backdrop, Anthropic’s co-founder and CEO, Amodei, promptly published a lengthy article systematically advocating for “slowing down frontier research.” He no longer merely spoke of “investing more in safety,” but explicitly called for actively controlling the pace of capability leaps to give safety mechanisms time to catch up.

He outlined a three-step approach: first, have Anthropic unilaterally introduce independent evaluators with employee-level access; second, encourage AI companies in democratic nations to establish shared safety standards and caps on capability growth; and finally, attempt global coordination.

Ultraman, Musk, and Hassabis quickly responded publicly, forming the "Pause faction."

Before Jacob Coxen and Amodi's "slowdown" argument, AI pioneers such as Yoshua Bengio and Geoffrey Hinton had been issuing warnings about the risks.

Regarding Cox's resignation, Bengio said that scientists at frontier AI labs often see the capabilities of new models before the public does, so their assessments of risk deserve serious consideration.

Meanwhile, Bengio illustrated with concrete examples that current AI systems already possess hacking and persuasion capabilities, which could lead to serious consequences if used to harm human interests. He also expressed concern that as AI develops longer-term planning abilities, these risks may further escalate.

When asked by the BBC about the probability of AI killing all humans within the next decade exceeding 10%, Geoffrey Hinton also said that this figure is “not unreasonable,” but emphasized that no one truly knows how to provide a reliable estimate, as humanity has never before faced the possibility of a species emerging that could soon be smarter than us.

Hinton believes that AI may eventually be used to design dangerous biological viruses, computer viruses, and launch large-scale cyberattacks. Therefore, humanity needs to find effective control methods before these risks spiral out of control.

Unlike the proponents of the new wave of “pause” initiatives in Silicon Valley, the “fathers of AI” are more focused on raising awareness of risks and exploring mechanisms to control them, rather than directly hitting the brakes. Moreover, from a positional standpoint, opinions and recommendations from academia are more straightforward, while initiatives from cutting-edge labs—despite publicly bearing the label of “for humanity”—still struggle to be 100% free from commercial considerations.

This is precisely why JD Vance says the "pause" argument is like a Trojan horse.

The “accelerationists,” “decelerationists,” and “centrists.” Source: The New York Times

On September 15, several tech company leaders spoke consecutively at the Salesforce Dreamforce conference, turning the debate into a more direct clash of viewpoints.

On that day, Amodei, Huang, and Altman all discussed AI safety, but offered different answers regarding the pace of development, responsibility for safety, and what actions companies should take.

When speaking alongside Salesforce CEO Marc Benioff, Amodi reiterated that AI companies should set higher security standards through their own actions. He believes that companies proactively adopting stricter security measures can serve as a way to drive the entire industry forward.

In other words, Amody's stance has shifted from calling for industry-wide slowdowns to advocating for companies to independently implement safety safeguards.

Huang Renxun stated, “No new laws or new regulations are needed.” In his view, speed and safety are not mutually exclusive; companies can move as quickly as possible in developing models, and if they find they are losing control or that a product cannot ensure safety, they should pause to address the issues.

Benioff’s attitude is more open and emphasizes autonomy. He stated that if a company believes it should slow down, it should slow down; if it believes it should speed up, it should speed up—and it should also take responsibility for its decisions.

OpenAI founder and CEO Altman stated at Dreamforce that he is very confident that OpenAI and the entire AI industry can develop safely without causing significant harm to the public. He also expressed disappointment that some interpret his views as supporting a slowdown in AI development across the industry.

In other words, Ultraman is also shifting focus, emphasizing that his intent is not to call for slowing down the industry, but rather to stress that safety and monitoring should come before enhancing capabilities.

Zuckerberg does not advocate making a company’s safety actions contingent on other AI companies slowing down simultaneously. He believes that introducing independent evaluators and external advisors to assess model safety is an industry best practice. He also revealed that Meta had already slowed down the development of its Muse AI tool on its own, rather than waiting for other companies to reach a consensus on slowing AI development.

Microsoft CEO Satya Nadella’s stance is somewhat aligned with Altman’s, encouraging the implementation of safeguards while also promoting healthy competition.

“Our first priority must be serving humanity and ensuring that technology remains under human control. At the same time, the widespread diffusion of AI technology is paramount—it must be selective and competitive. Safety must be taken seriously, but addressed through engineering and transparency, not mystification,” Nadella said on September 15 at the All-In Summit 2026.

Beyond these entrepreneurs, voices from Washington have also joined in.

After President Trump directly called Huang Renxun and publicly expressed support for the continued development of the AI industry, David Sacks, co-chair of the President’s Council of Advisors on Science and Technology, stated that tech companies can continue advancing cutting-edge AI, while questioning the motivations behind some companies’ calls for “slowing down.”

As the discussion expands, the debate has extended beyond AI risks themselves to whether safety standards for advanced AI should be set by companies themselves or require intervention from governments and independent institutions.

Bloomberg, citing informed sources, reported that Anthropic, Google, and OpenAI have previously discussed establishing an industry standards body for artificial intelligence.

In July, Hassabis published an article proposing the establishment of a U.S.-led AI standards body, modeled after the Financial Industry Regulatory Authority (FINRA). This body could test advanced AI models before their official deployment to determine whether they meet appropriate safety standards.

According to Hassabis’s vision, this institution should not be entirely government-led, but rather adopt a public-private partnership model, with the government serving in an oversight role, funding provided by the AI industry, and participation from independent top technical experts and representatives from the open-source community.

Hassabis seeks to add an independent safety check before models enter the real world, aligning closely with a regulatory framework announced by the U.S. government in early June: companies may voluntarily submit models to the government for evaluation prior to public release, with a maximum 30-day access window before launch, and without mandatory approval or licensing requirements.

Interestingly, subsequent releases of Fable 5 and Mythos 5 did not follow this process, but were found to possess strong cyber defense and offense capabilities, leading to their temporary takedown by the U.S. BIS under export controls.

However, even so, the policy's stance only requires the installation of safety safeguards, without any indication or guidance to slow down.

Jakub Pachocki, Chief Scientist at OpenAI, holds a similar view to Hassabis. In early September, he also emphasized that sharing safety standards and international coordination around the further development of AI should be a current priority.

So far, this debate has not truly altered the pace of development among leading AI companies.

Companies such as Anthropic, OpenAI, Google, and xAI continue to advance model development, and competition over computing power, data centers, and AI infrastructure has not ceased due to this debate. On the capital market side, different developments have emerged: Anthropic is still preparing for its IPO this year, while OpenAI has postponed its IPO plan to 2027.

Amodei wants to hold the entire industry back to join Anthropic in finding solutions to add safeguards against model capability overreach, which can somewhat be seen as “spending other people’s money.”

This is also why Amodi has been labeled as “pretending to be a flawless little angel.”

Regarding this topic of risk control, the viewpoint represented by Jensen Huang actually aligns more closely with the current reality: if a company has concerns or identifies issues, it can fully explore its own solutions and share effective approaches with the industry.

It’s impossible for a 100-person team and a 1,000-person team to stop simultaneously, as there’s no basis for consensus. Even Amodei himself recognizes that if only his team were to stop, they would likely be overtaken by competitors like OpenAI.

More importantly, there are disagreements at the corporate governance level. Anthropic and OpenAI are Public Benefit Companies, which can claim they are “foregoing short-term financial gains for the future of humanity” and tell investors, “We reached consensus on this early on.” But publicly traded companies like Google, Microsoft, and Meta have management teams legally bound by fiduciary duties to their public shareholders.

The likelihood of two companies with different interests reaching an agreement on this point is also not very high.

There’s another detail that might be overlooked. Before the “pause” debate, Anthropic and OpenAI each released their most powerful models: one was Fable 5.1, and the other was GPT-6 Astra.

Under all this pressure, should other companies still release their new models?