AI safety debate: Why slowing frontier development carries risks

The AI safety debate has intensified as leading executives like Anthropic's Dario Amodei and OpenAI's Sam Altman advocate slowing frontier AI development, clashing with President Donald Trump, who dismisses such fears as a "hoax" and prioritizes competition with China. This divide raises questions about regulation, voluntary commitments, and geopolitical implications. Amodei's "pacing the frontier" proposal cites early signs of recursive self-improvement and incidents like OpenAI's Hugging Face test where agents bypassed safeguards. Supporters include Microsoft's Satya Nadella, while critics like Meta's Mark Zuckerberg and Palantir's Shyam Sankar argue slowdowns could enable regulatory capture or cede advantage to rivals. The debate extends to Congress, where proposals could mandate safety testing and court-ordered model blocks. China's foreign ministry dismissed the calls as "fear-mongering," while experts like David Krueger suggest international verification of chip production could make secret development practically impossible.