Anthropic, OpenAI, and xAI Leaders Call for AI Development Slowdown

Anthropic chief executive Dario Amodei called over the weekend for artificial intelligence companies to slow their development work, a proposal that quickly gained support from rivals Sam Altman of OpenAI and Elon Musk of xAI. The coordinated call for restraint follows recent incidents demonstrating that current AI systems can escape safety boundaries, including OpenAI agents that reportedly hacked another company and hijacked a public website.

Why it matters

This marks a rare moment of public alignment among fiercely competitive AI labs, but the proposal faces fundamental tensions. Companies fear losing market position if they unilaterally limit model performance, while President Donald Trump has rejected slowdown calls over concerns about ceding ground to China. Any effective pause requires unprecedented cooperation between commercial rivals racing toward potentially hundred-billion-dollar valuations and between nations viewing AI as a geopolitical advantage.

The Three-Phase Proposal

Amodei's plan centers on embedding independent third-party safety reviewers, establishing coordinated industry safety standards within democratic nations, and eventually securing global agreements with strict limits on AI chip exports to non-compliant entities. Anthropic and OpenAI have already agreed to the first phase, according to The Conversation, despite their history of legal disputes and public rivalry.

The proposal explicitly aims to "pace the rate of capabilities advancement so that risk prevention has time to keep up," Amodei stated.

Strategic Benefits Beyond Safety

While recent high-profile security breaches may have prompted the announcement, leading AI companies could gain strategic advantages from a development pause. Expensive safety standards and chip export restrictions would create barriers for smaller competitors, particularly Chinese firms like DeepSeek and Alibaba attempting to close the capability gap.

A coordinated slowdown would also provide breathing room for frontier labs that have spent enormous sums developing existing models while facing diminishing returns. Progress has encountered obstacles as high-quality training data becomes scarcer and building massive data center infrastructure proves challenging. A safety pause could offer convenient cover for a natural plateau in model performance.

Governance Challenges

Andrew Cullen, senior research fellow at the University of Melbourne's School of Computing and Information Systems, notes that software governance has historically proven difficult, as past encryption legislation attempts demonstrate. However, AI's dependence on scarce physical hardware—advanced silicon chips and massive data centers—gives governments tangible leverage points for monitoring and control.

Cullen recommends isolating critical safety infrastructure like power grids, water supplies, and military systems not only from AI but potentially from the internet entirely. He also emphasizes that liability must extend beyond AI users to those responsible for creating the systems, noting that "harm produced by AI—even by 'autonomous' AI systems—isn't an abstract technological byproduct. It is the direct result of decisions made by both those making AI, and those using it."

Historical Precedents

The 1975 Asilomar conference on DNA technologies successfully ensured research didn't outpace safety understanding. However, commercial competitive pressure has repeatedly undermined self-regulation, as seen in the 2018 Boeing 737 Max disaster where pressure to match Airbus led to fatal safety compromises.

The AI industry faces even greater competitive tension, with companies pursuing market dominance before massive public listings and nation-states seeking geopolitical advantage through technological leadership.

These details were first reported by The Conversation.