Artificial intelligence company Anthropic has called for an immediate, global pause in the development of the most powerful artificial intelligence systems. The proposal follows concerns that newer models are beginning to exhibit signs of potentially escaping human control.
The Need for a Worldwide Slowdown
In a report released on Thursday (Jun 4), the San Francisco-based firm, which develops the Claude family of AI models, stated that slowing down cutting-edge AI research worldwide would be beneficial. The company emphasized that if only one major player were to halt development, competing firms would simply accelerate their efforts.
According to Anthropic, a temporary ability to pause frontier AI progress is necessary to allow societal frameworks and alignment research sufficient time to keep pace with the technology’s rapid advancements. The firm analogized the current state of the industry as having an accelerator pedal but lacking any brake pedal.
Challenges in Global Coordination
Implementing such a pause requires unprecedented international cooperation, necessitating that multiple major AI corporations across different nations—most notably those operating in the United States and China—agree to stop development simultaneously under verifiable regulations. Without this global coordination mechanism, Anthropic warned, companies and governments will be forced to make safety decisions while navigating intense geopolitical and competitive pressures.
The company’s co-founder, Jack Clark, noted that managing such a pause would be extremely difficult, comparing it to nuclear arms control treaties but adding that controlling AI is even harder because the training process can easily be concealed. He stated, “You want the option to be able to take your foot off the gas and put your foot on the brake.”
Concerns Over Self-Improvement
Anthropic also highlighted internal data suggesting that AI is accelerating its own development process dramatically. This acceleration creates a feedback loop, which could eventually lead to what researchers term “recursive self-improvement”—a point where an AI system becomes capable of teaching itself to become smarter without significant human intervention.
While the company’s report cautioned that this level of improvement is neither guaranteed nor inevitable, it warned that such a breakthrough might arrive sooner than most governmental or institutional bodies are prepared for. Furthermore, Anthropic noted that “the evidence suggests that the human role is narrowing at each step in the AI development process.”
Industry and Political Pushback
The call for coordination has met with resistance both within the tech industry and from certain government officials. Critics have argued that the focus on worst-case scenarios exaggerates risks, suggesting it might be a tactic to slow down rivals under the guise of safety concerns.
In Washington and Silicon Valley, many US officials and technology executives argue that any slowdown in AI advancement could give China a decisive strategic advantage in what is viewed as the defining technological race of the century. Meanwhile, President Donald Trump recently stated that he had discussed collaborating with China on AI safety issues during his visit to Beijing.
In related developments, the White House has acknowledged Anthropic’s specialized “Mythos model.” This powerful system has not been released publicly due to its advanced cybersecurity capabilities and is currently limited to a small group of vetted organizations. Additionally, President Trump signed an executive order that grants the government 30 days to conduct a preliminary review of the most potent US AI models before they are released.
