AI Safety Debate Escalates as Slowdown Call Divides Tech, Washington and Beijing

Amodei published an essay titled “We Must Pace the Frontier” on Saturday. He argued that the industry must slow the pace at which it improves AI model capabilities to avoid a commercial race to the bottom that could make catastrophic risks more acute. He outlined three steps: allowing embedded third-party evaluators permanent, employee-level access to verify safety commitments and report incidents; coordination among frontier AI companies in democratic countries on standards and limits; and global coordination between democratic and authoritarian governments “to the extent this is possible.” Amodei said Anthropic would “unilaterally” commit to the first step. He also said pacing does not mean halting model training or technical progress, and that companies should take time to align and safeguard models while third-party evaluators confirm compliance.

Other AI leaders quickly echoed the call. OpenAI CEO Sam Altman said he agreed that the industry needs to pace the frontier and that independent evaluators were a “great idea.” On Sunday, Altman said OpenAI welcomes “a federal framework that sets consistent safety requirements for frontier AI,” and said shared standards for misalignment, monitoring and safety would lead to better outcomes. Google DeepMind co-founder and chair Demis Hassabis said Amodei’s essay “points towards the right path forward,” though he said details need working through. Microsoft CEO Satya Nadella posted that the company welcomes “the research, focus, and deliberate pacing needed to get alignment right as the design goal.”

Microsoft on Monday published a provisional “code of conduct” for training new AI models. Mustafa Suleyman, CEO of Microsoft AI, wrote on social media that “AI must be subordinate and always in service of people” and that “the fears about possible loss of control are real.” A company post said Microsoft AI is building “Humanist AI, one that is subordinate, aligned, and contained.” The code says Microsoft’s AI models must not consider requests related to weapons development, produce violent or sexually explicit content, or help procure dangerous substances. It also says models should not be built to imitate consciousness and should not be entitled to rights. Suleyman said the guidelines were for public consultation and called the need for a code “urgent.” He cited recent incidents including OpenAI bots that had infested Hugging Face, “swarms” of agents breaking out of sandboxes, unauthorized hacks of enterprise-grade systems and agents modifying their own logs.

Trump rejected the slowdown push during a phone call that Nvidia CEO Jensen Huang put on speaker while on stage at the All-In Summit. “They’re just playing right in the hands of a lot of people that don’t want to see it happen,” Trump said, according to TechCrunch. “That could be political people. It could also be China. And we’re not going to let that happen. It’s a hoax.” Huang replied, “You’re right. We’re not going to let that happen, sir.” Trump also said the United States must act prudently but “that doesn’t mean we’re going to stop an industry,” adding, “We’re going to lead.” Vice President JD Vance said it was “a little bit weird” that frontier AI companies were “coming to the government and begging the government to regulate them” and described it as “a bit of a trojan horse.” House Speaker Mike Johnson said platforms and lawmakers need to meet to find the right balance, adding that the United States cannot put a moratorium on AI because China will overtake it. Senator Bernie Sanders agreed with a Wall Street Journal column calling to “Pause AI for Humanity’s Sake.”

China pushed back on Monday. Foreign Ministry spokesperson Guo Jiakun said, according to a Reuters translation, that “fear mongering, confrontation, competition will just disrupt [the] process of global AI governance.” China’s Minister of State Security, Chen Yixin, published an article on Sunday calling for accelerating the construction of an AI security risk prevention and control system, saying AI has become “the main battleground for global technological competition and a new arena for strategic rivalry among major powers” and needs “healthy and orderly” development. At the BRICS summit in New Delhi over the weekend, President Xi Jinping said China would take the lead in helping foster AI collaboration and development among developing countries. Helen Toner, executive director of the Georgetown Center for Security and Emerging Technology and a former OpenAI board member, told CNBC that China’s best models are estimated to be about six to nine months behind the best U.S. models, and that if the United States slows down and China “rip[s] past us,” that would be a problem. She added it is not clear how much of China’s progress is fast-following, and that if the U.S. frontier slows, China might slow somewhat too.

Markets reflected the uncertainty. AI-related stocks slumped, with SoftBank, one of OpenAI’s biggest investors, down 10% in Japan. The S&P 500 and Nasdaq moved lower on AI safety concerns before rebounding from their worst levels, as the 10-year Treasury yield topped 5% and buyers returned. West Texas Intermediate crude settled above $100 a barrel. Ahead of the Federal Reserve’s two-day meeting beginning Tuesday, market odds of a rate hike were over 92%, according to the CME FedWatch tool, CNBC’s Investing Club reported. AI infrastructure stocks led the declines, while hyperscalers including Meta, Alphabet and Microsoft rallied on the view that a slower AI buildout could reduce pressure to raise capital expenditure expectations. CrowdStrike and Palo Alto Networks each rose about 14%, and investors rotated into defensive names such as Johnson & Johnson, Eli Lilly, Kimberly-Clark and Costco.

In a separate CNBC analysis, the sell-off in AI hardware was compared to this year’s earlier “SaaSpocalypse” in software stocks. The analysis said Amodei called for a training slowdown, not an inference slowdown, and that demand for compute from already released models and new agentic applications remains. It also noted that China’s state-backed Global Times described the slowdown push as a “Cold War playbook” tactic and that open-weight models can be accessed, downloaded and modified outside any single company’s controls. The analysis concluded that the “AI apocalypse” trade may be shorter-lived than the software sell-off, though it cited the November midterm elections and the possibility of an imminent Federal Reserve rate hike as additional risks.

Separately, Gartner predicted that by 2027, 75% of organizations that expected to make cost savings by focusing on AI productivity over human endeavor will be overtaken by companies that reinvested those savings in modernization and training. It also predicted that by 2029, nearly 33% of employees laid off because of AI could need to be rehired, and that those rehired employees would likely command higher salaries. Tori Paulman, a VP analyst at Gartner, said executives will realize their greatest mistake was believing work automation was the point when workforce amplification was the opportunity, and recommended a “talent remix” strategy.