AI Safety Alert: Global Experts Call for Pause on Giant Model Training

The Future of Life Institute released an open letter on March 29, 2023, signed by over 3,000 individuals including Elon Musk and Yann LeCun, calling for an immediate, global 6-month pause on training AI systems more powerful than GPT-4. The letter urges AI labs to halt the uncontrolled race toward increasingly powerful digital minds and instead prioritize safety and governance.

Safety Concerns：The Unchecked Race in AI Development

The letter argues that contemporary AI systems are now reaching human-competitive performance on general tasks, yet their rapid deployment lacks adequate safety oversight. Key risks highlighted include AI flooding information channels with propaganda and falsehoods, automation displacing human workers—including fulfilling jobs—and the potential development of nonhuman intelligence that could eventually outnumber, outsmart, or replace humans. The document cites the widely endorsed Asilomar AI Principles, warning that advanced AI represents “a profound change in the history of life on Earth” requiring commensurate planning and resources—which is currently missing. Crucially, the letter notes that recent months have seen AI labs locked in an out-of-control race where models become so complex that even their creators cannot understand, predict, or reliably control them.

Governance and Regulation: Building New Frameworks

The letter proposes concrete actions: first, the pause must be public and verifiable, involving all key actors; if未能 rapid voluntary pause occurs, governments should institute a moratorium. Second, AI labs and independent experts should jointly develop and implement shared safety protocols that are rigorously audited by external experts to ensure safety “beyond a reasonable doubt.” Third, the letter urges dramatic acceleration in building robust AI governance, specifically recommending: new regulatory authorities dedicated to AI; oversight and tracking of highly capable AI systems and large computational resources; provenance and watermarking systems to distinguish real from synthetic content; an auditing and certification ecosystem; liability for AI-caused harm; robust public funding for AI safety research; and institutions prepared to handle AI-driven economic and political disruptions.

Balanced Progress: Optimizing Existing Systems

Importantly, the letter clarifies that the pause is not a halt to all AI development—it specifically targets the dangerous race toward ever-larger, unpredictable black-box models with emergent capabilities. Instead, research should refocus on making today’s state-of-the-art systems more accurate, safe, interpretable, transparent, robust, aligned, trustworthy, and loyal. This approach balances innovation with responsibility, allowing society time to adapt while enjoying existing AI advancements.

Recommendations for Practitioners

• Developers and enterprises currently deploying frontier models should assess risk exposure and prioritize already vetted, safer models rather than upgrading to unvalidated super-models.
• Organizations and policymakers should accelerate internal AI governance policies and contingency plans and monitor regulatory developments to prepare for incoming frameworks.
• Researchers may use this window to engage in safety protocol discussions or pivot toward interpretability, robustness, and alignment research.

In Conclusion

This letter expresses responsible optimism about AI’s potential, not technological pessimism. Just as humanity paused high-risk technologies like human cloning and gene editing to establish safeguards, the AI field needs同样的 prudence—the goal is a sustainable, long “AI summer,” not a rushed plunge into autumn.