In recent days, several CEOs of competing artificial intelligence companies have publicly supported slowing the development pace of the most advanced models. The discussion involves Dario Amodei (Anthropic), Sam Altman (OpenAI), Demis Hassabis (Google DeepMind) and Elon Musk (xAI).
Dario Amodei published a proposal titled We Must Pace the Frontier, arguing that model capabilities are improving so rapidly that safety systems and oversight cannot keep up. He called for independent external auditors with deep access to models and urged shared industry and international safety standards; Anthropic said it would unilaterally take initial steps. Sam Altman of OpenAI signaled last week he is open to slowing development, preferably in coordination with other leading firms, and publicly agreed with Amodei’s suggestions. Elon Musk posted on X, “Dario is right.” Demis Hassabis also expressed support.
Why slow down? Official reasons and underlying incentives
Officially, the call to slow development centers on safety and oversight. Amodei warns that autonomous AI agents are increasingly capable of complex tasks without human intervention and could eventually enable serious cyberattacks, system breaches or persistent botnets. Slower development would provide time for more extensive testing and for building independent audit mechanisms.
The proposals also emphasize growing legal and financial risks: more powerful and autonomous models could cause greater harm in the event of incidents, raising product liability and compliance questions.
Beyond safety, several economic and strategic motives may be at play:
- Financing pressure: advancing models requires ever more chips, data-center capacity, energy and capital, while fundraising conditions have tightened.
- Time to monetize: slower development could allow companies to deploy and commercialize existing models before investing in the next leap.
- Regulatory influence: if leading firms set audits and standards, they may shape future government regulation to their advantage.
- Market entrenchment risk: costly audit and safety requirements could become de facto entry barriers that favor large incumbents over startups.
Some observers, including investors, suggest that safety messaging might partly mask natural technological slowdowns or strategic economic aims.
Historical and industry parallels
The article notes that voluntary technological restraint has precedents. In the 1970s recombinant DNA researchers called for temporary moratoria and later established safety guidelines at the Asilomar conference. The auto industry has self-imposed limits on performance in various eras, and Formula 1 teams and engine manufacturers agreed to freeze certain developments. These examples show that competitive sectors have sometimes accepted self-restraint for safety, regulatory or commercial reasons.
OPEC analogy — how far does it hold?
The piece compares the proposal to OPEC-style coordination: oil producers limit output to support prices, reducing incentives for one actor to flood the market. Likewise, coordinated AI slowdown could reduce an arms race. Important differences remain:
- Observability: oil production is measurable; AI development speed and secret advances are far harder to verify.
- Stakes: AI development entails not only market share but also geopolitical power.
- Incentives to defect: a single actor can gain a decisive advantage by continuing development while others slow.
The biggest complication: China
A central obstacle to global coordination is China. Amodei envisions a phased approach—company-level audits, then shared standards among democratic countries, and finally global coordination including China and other states. However, if China does not participate, the scheme may fail.
China has already replied: the Chinese Ministry of Foreign Affairs warned that fearmongering, confrontation and malign competition hinder global AI governance. Beijing argues that AI risks should be addressed through international cooperation and shared regulation rather than by slowing technological progress or singling out China.
Outlook: limited odds, high stakes
Although the initiative is notable because competing firms rarely agree to self-limits, practical challenges remain: verification, incentives to cheat, and geopolitical rivalry. The ultimate effect will depend on whether international agreement can be achieved and whether leading companies will accept the transparency and costs that meaningful audits and standards require.
Related events
The media outlet mentioned is organizing industry events: an Investment Day on October 21 and an AI & Digital Transformation conference on November 26, which will address investment and technological implications related to these topics.
This article does not constitute investment advice or recommendation.
