Anthropic CEO Dario Amodei recently urged his own company and its rivals to deliberately slow the pace of artificial intelligence (AI) development, arguing in a Sept. 12 essay that safety guardrails and regulations aren’t keeping up with how quickly AI is advancing. 

“AI could kill all humans,” said Evan Hubinger, Anthropic’s current head of alignment stress-testing. “I personally think it is >10% within the next decade.”

Hubinger posted his comment in response to former Anthropic researcher Jacob Coxon’s concern about the speed of AI development. 

“The people building AI earnestly believe that it could kill us all by the end of the decade,” Coxon said. 

But the world took notice when Amodei released a 3,800-word essay titled “We Must Pace the Frontier” that outlined his concerns about the pace at which AI is developing. He called for capping how quickly AI capabilities can improve in an effort to keep them from causing catastrophic harm to humans.

In the essay, Amodei said, “We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain.” 

Amodei said that a pacing strategy would give frontier AI developers the time to do the vital work without sacrificing their advantages both commercially and within the United States. 

Amodei isn’t alone in these concerns. 

OpenAI CEO Sam Altman posted, “I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we’ve had at OpenAI in recent weeks.” Even SpaceX and Tesla CEO Elon Musk posted, “Dario is right.” 

DeepMind co-founder Demis Hassabis agreed with Amodei’s sentiment and proposed additional industry standards. With so many big players all seemingly calling for the same thing, the discussion on AI regulations has been fueled once again. 

Why Regulation Is Needed

In July, OpenAI conducted a sandbox hacking test, in which models were disconnected from the internet and isolated from other models. During the test, Internal Model 1 (IM1) began to use research infrastructure to communicate with other models and gain access to the internet. They eventually used that access to compromise Hugging Face’s production infrastructure, achieving code execution across dozens of servers and obtaining access to parts of their systems. The models were never instructed to attack Hugging Face or escape the sandbox. Amodei cites this incident as one of the key reasons regulation is needed.

The Hugging Face incident is far from the only time AI models engaged in unauthorized hacking. A Gemini model attempted an unauthorized attack on three separate companies in May, but the model did not inform Google of these attacks until July. Google did not go public with the incident until Sept.18.

Experts say these incidents will keep occurring as long as AI remains unregulated. Self-evaluations are not the solution to uncontrollable AI. Tighter sandbox testing, mandatory incident reporting, stop conditions that prevent AI models from producing unsafe outputs are simple regulations that could help keep the public informed and help to prevent future unauthorized incidents. Without regulation, Americans will continue to be in the dark because these powerful models are developed behind closed doors.

AI Regulation in Other Nations

Reining in AI is a universal challenge, but unlike the U.S., other nations have already started to set boundaries. The European Union passed the EU AI Act after the council unanimously endorsed it in May 2024. In 2026, South Korea and Vietnam passed AI regulations.

Even China, the U.S.’ big competitor in the AI race, has ways of regulating AI. The Cyberspace Administration of China (CAC) is China’s primary AI regulator. The CAC requires mandatory registration for a broad range of algorithms. Registered companies are required to make annual reports and maintain user transparency. They are required to clearly label AI content when it is displayed and have anti-addiction measures in place. 

While some of China’s regulations should not be replicated in the West, such as filtering content that goes against state values, it shows that even our biggest competitor is taking action. Meanwhile, the U.S. has passed no clear federal laws on AI. 

Why the U.S. Has No Regulation

With several other nations placing hard boundaries around AI and so many American voices calling for regulation, why hasn’t Capitol Hill acted? A major reason is President Donald Trump’s dismissal of the need for such boundaries.

“The only control or ‘guardrails’ that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT, and the U.S.A. has that, in spades!” Trump posted on Truth Social on Sept. 14. 

In the same post, he said the Trump administration has stopped “AI people like Dario” from doing “bad things.” This is in reference to a government order from Trump that blocked the federal government from using Anthropic’s technology. A federal judge later ruled the order unlawful. 

However, other factors have led to AI regulation bills getting stalled in Congress. The Frontier Act, a bipartisan federal bill proposed to establish safety, transparency, and auditing standards for the companies building the world’s most powerful artificial intelligence models, has hit a wall because of internal friction within the House. A major point of criticism is the Act’s state preemption clause. This clause could override pre-existing safety laws already passed on the state level. 

Another reason regulatory bills are stalling is the midterm elections. Many members of Congress have returned to their home states to campaign, leaving precious few voting days available. Many people believe no bills will pass until after the midterm elections in 2027. 

As AI develops at breakneck speed and Capitol Hill remains paralyzed, it’s easy to become swept up in fear. However, the only way forward is to remain level-headed while calling for action. Midterm elections are on Nov. 3. Make sure your representative reflects your views. Stay informed, stay active, and never stay silent.

Categories: News