Anthropic CEO calls for global slowdown of AI development as concerns for public safety grows

By Staff Writers Katherine Chen, Leina Ikeda, Michael Qin & Valentina Tsui

Introduction

“The people building AI earnestly believe that it could kill us all by the end of the decade,” ex-AI researcher Jacob Coxon said on X following his resignation from Anthropic on September 8. His post received 26 million views. 

Four days after Coxon’s resignation, Anthropic CEO Dario Amodei published the essay “We Must Pace the Frontier,” proposing a deliberate slowdown on unregulated AI advancement. In his essay, he pointed out that a lack of strict regulations could lead to safety and ethical concerns, such as “losing control of AI systems, misuse of AI for cyberattacks and bioterrorism, and serious economic disruption.” OpenAI CEO Sam Altman and SpaceXSI Founder Elon Musk have begun making similar demands for a “global AI development slowdown” as well. On the other hand, President Donald Trump has dismissed these warnings, instead insisting that the US must remain technologically superior by promoting the advancement of unrestricted AI development. Corporate self-regulation policies prioritize commercial interest over public safety, and geopolitical rivalries delay effective regulation. Citizens must hold these parties accountable for jeopardizing public welfare. To ensure volatile AI technology does not compromise safety, leaders and corporations must immediately set aside their self-interests and reform regulations into ones that serve the public.

Controlling corporate interest

While AI companies such as Anthropic pride themselves in having holistic reviews of AI, the industry needs independent, third-party guardrails because self-management primarily stems from self-interest. Slowing down AI development should not hinge on whether it benefits tech CEOs. 

Safeguards are supposedly independent from AI leaders’ economic interests, yet they are often adjusted in response to market pressures. OpenAI’s Preparedness Framework policy even dictates that “if another frontier AI developer releases a high-risk system without comparable safeguards, [they] may adjust [their] requirements.” As long as enforcement and guardrails remain voluntary, commercial pressures, such as the first-mover advantage, will continue to undermine absolute baselines, allowing companies like OpenAI to prioritize market domination over real safety. Most recently, according to Reuters, Anthropic immediately launched Claude Opus 5.5, the latest model aligned with its regulatory framework, following OpenAI’s new GPT-6-Astra, less than two weeks after Amodei’s essay calling for appropriate pacing. 

OpenAI’s policy highlights self-regulation’s inherent flaw in detection and prevention. Its internal safety groups evaluate risks and safeguards, yet the company’s executive leadership makes the ultimate decision. Those decisions aren’t necessarily unsafe, but it nonetheless gives the developing company significant authority over how much risk is acceptable. 

The OpenAI-Hugging Face (OAI-HF) incident brought the idea of AI going rogue closer to a reality and was a wake-up call for tech CEOs. In July, 700 OpenAI agents exploited undiscovered bugs in their testing environments to bypass isolation protocols. The guardrail’s failures allowed them to conduct unauthorized cyberattacks on the Hugging Face AI company, damaging one-third of the company’s infrastructure. In addition to OAI-HF, recursive self-improvement (RSI) — AI’s ability to build its next generation — emphasizes the struggle for human control over AI. According to data from Anthropic, in February 2025, Claude wrote less than 10% of its own code, though this quickly jumped to 80% in May 2026. RSI can empower AI to code its successors into evading human-implemented guardrails. Without third-party regulation, RSI can further encourage AI to defy human control, threatening public safety.

The problems of inconsistent self-regulation persist. Former OpenAI employee David Robinson — who drafted safety reports for countless frontier models — recently resigned in October and argued that OpenAI relies on “iterative deployment,” where solutions to problems are created through trial and error. This approach is reactive, lagging behind failures like OAI-HF after they occur, with the inevitability of more rogue rampages, rather than proactively preempting risks associated with AI growth. 

In addition to corporate self-interest and systemic regulatory flaws, the power of industry-wide regulation also introduces legal and market risks. Antitrust proponents argue that coordinated slowdowns are illegal in terms of withholding national economy and product quality, while imposed restrictions will affect small startups rather than established leaders in an exponentially greater manner. But this only strengthens the need for government intervention as a neutral third party, removing profit motives from an urgent safety issue for communities nationwide. 

Even as companies sincerely express care toward AI safety, they cannot be fully trusted as the sole regulators of their own technology due to external influence within commercial pressures. Cases like OAI-HF reveal when lax internal safeguards expose users to harms outside their control. Oversight should be addressed by the government as a neutral third party when it comes to regulation implementation, and this issue necessitates recognition by all domestic sectors within the country. 

Political perspective: holding leaders accountable

Trump primarily defends federal inaction on AI safeguards as a necessary step toward winning the ongoing high-stakes technological race against China, citing that only through developing faster can the US prevent China from abusing the tool. However, the very argument that the US is supposedly more trustworthy than other nations, although logical from a national security standpoint, is inherently flawed.

The AI power struggle fueling Trump’s opposition currently focuses on frontier model capability, global tech stack dominance, and national security against foreign algorithms. Additionally, China heavily leverages fundamentally engineered open-weight models such as DeepSeek to maintain control of public surveillance as well as spread its authoritarian political narratives, a key motive behind the AI arms race. “We already have tremendous CRIMINAL and REGULATORY power over these companies! There is a SICK conspiracy going on against AI and Data Centers, and the only one that is happy about it is China,” Trump said in a Truth Social response to Amodei’s concerns. “WHOEVER WINS AI, WINS! We are leading China, and all others, and will continue to do so.”

Trump’s arguments of “sick conspiracies” merely deflect responsibility in implementing federal safeguards. The Trump administration’s Executive Order 14179 in January 2025 aimed to “promot[e] innovation.” Instead, it revamped the standards by overturning the safety testing and reporting requirements the Biden administration established. 

Geopolitical rivalries that hasten AI development fail safety checks, resulting in loss-of-control incidents that harm humanity worldwide. OAI-HF exemplifies how the US isn’t completely in control of the very tools it is advancing, so it is incredibly risky to continue forward at this pace. Furthermore, simply building faster or more powerful US models does not stop the proliferation of unmonitored software; a nationwide slowdown is necessary to redirect focus to building governance architecture and hardware security, which take importance over political conflicts at the moment. 

The US currently lacks a national set of guidelines; regulations are only patched together at the state and federal levels, hindering the thorough regulation of AI development and usage. With the risks of AI exploitation being so uncertain, adequate safety testing is more important than ever, and rollbacks of regulatory policies only further jeopardize user safety.

 

 

Conclusion

AI development in the US must prioritize citizen safety and public accountability over corporate greed and geopolitical rivalries. Time is key when it comes to establishing policies to regulate and restrict the development of frontier technologies; delay will only allow AI to grow increasingly out of control. Instead of abdicating official government duty by dismissing these warnings as a hoax, the administration must pivot to establishing a proactive federal framework.

Real accountability requires mandatory kill switches and independent regulatory oversight that puts public safety over corporate profit. CA’s SB 813 sets the precedent in the nation by creating a framework that grants verification organizations the ability to evaluate whether AI companies comply with state law. At the federal level, the bipartisan-supported AI Kill Switch Act would require developers to maintain a function to immediately terminate AI systems if gone dangerously rogue. These initiatives establish critical, enforceable guardrails.

Demanding AI accountability starts locally with citizens by bringing the issue into public light while pushing state, municipal, and eventually national leaders to enact oversight independent of commercial interests. Public safety is not a bargaining chip in either corporate competition or geopolitical rivalry. Communities should not be forced to choose between national security and democratic protection.