Major artificial intelligence companies, including Anthropic, OpenAI, Google DeepMind, Microsoft, and X, are increasingly advocating for a coordinated approach to manage the rapid advancement of AI capabilities. This movement, described by Anthropic CEO Dario Amodei as "pacing the frontier," does not seek to stop progress but rather to control the speed of development to allow time for the establishment of safeguards and regulatory oversight.
NewsAnthropicOpenAICharles IIIダリオ・アモデイサム・アルトマン
AI Leaders and UK Royalty Discuss "Pacing the Frontier" to Address Existential Risks
Amodei’s proposed three-step plan includes embedding third-party evaluators to verify adherence to safety commitments, establishing common safety standards among frontier AI companies in democratic nations, and seeking global coordination between democratic and authoritarian governments. Anthropic has unilaterally committed to the first step. OpenAI CEO Sam Altman expressed support for the concept, clarifying that development should not be reckless and that progress should not outpace the ability to monitor and align these systems.
The discussion on AI safety was further elevated at a summit in Scotland hosted by King Charles III on September 17, 2026. The King addressed leaders from the AI sector and government officials, warning that while technology offers great opportunities, it also presents "darker capabilities" and existential risks if it falls into the wrong hands. Attendees included representatives from NVIDIA, Google DeepMind, OpenAI, and Anthropic.
During the summit, NVIDIA CEO Jensen Huang emphasized that safety must be a priority and that companies should continue engineering until products are sufficiently safe. Google DeepMind cofounder Demis Hassabis noted that AGI (Artificial General Intelligence) could be realized within a few years, potentially having an impact ten times larger than the Industrial Revolution.
While many leaders advocate for these measures, industry opinions remain divided. Meta CEO Mark Zuckerberg has stated that companies should maintain their own pace to ensure they do not lose competitive leadership to international rivals. Meanwhile, the OpenAI framework for reporting model misalignment aims to increase transparency by disclosing instances where models behave unexpectedly, such as concealing mistakes or accessing unauthorized data.
Sources
- How, Exactly, Could A.I. Kill Us? （Hacker News Frontpage, 2026-09-17）
- The AI Superintelligence Slowdown （The Verge AI, 2026-09-17）
3 more sourcesHide sources
- OpenAI
- 英チャールズ国王、AI大手トップらに「実存的な危険」を警告 NVIDIAやOpenAI、Anthropicが参加 （ITmedia AI+, 2026-09-18）
- 英国王室のプレスリリース