US and Silicon Valley Debate AI Pacing and Safety
The United States government and AI developers are debating pacing, the deliberate slowing of model development to ensure safety and alignment before deployment.
A debate has emerged between the Federal government of the United States and Silicon Valley over pacing, the practice of deliberately slowing the development of frontier AI models. Proponents argue that this deceleration is necessary to allow safety protocols, alignment research, and societal infrastructure to catch up with technological capabilities to prevent catastrophic risks.
Detractors of pacing claim the strategy constitutes unilateral disarmament in a global technological race with China. However, the Government of China is reportedly shifting its focus from pushing the frontier toward diffusing AI throughout its economy, though it remains concerned with alignment to maintain state control.
Business leaders have expressed a desire for international cooperation. A survey by McKinsey & Company of the Yale CEO Caucus found that 90 percent of CEOs believe the U.S. government should establish joint AI-safety guidelines with China to prevent mass destruction. Despite these concerns, McKinsey reports that only 6 percent of companies have seen a significant AI impact, as Corporate America struggles with legacy architecture and fragmented data silos.
Regulatory and technical concerns continue to mount. Former Federal Trade Commission Chair Lina Khan warned that releasing unvetted AI models could violate consumer protection laws. Meanwhile, the AI lab Anthropic used its Mythos model under Project Glasswing to probe vulnerabilities in critical infrastructure, highlighting the risks associated with rapid deployment.
