AI Leaders Propose Slowdown, But Implementation Faces Obstacles

Industry consensus emerges on AI safety

Leading artificial intelligence companies have reached a rare agreement: AI development should slow down to allow safety measures to catch up. Anthropic CEO Dario Amodei, OpenAI's Sam Altman, and other prominent figures voiced support for pacing AI progress after an Anthropic researcher resigned with warnings about threats to humanity.

The convergence comes at a critical moment as both companies prepare for potential public offerings and face pressure to maintain competitive advantages over rivals, particularly Chinese AI firms.

Why it matters

This marks the first significant industry-wide acknowledgment that the race to build more powerful AI systems may be outpacing the ability to ensure safety. Unlike past debates over AI risk, major commercial players are now proposing concrete mechanisms—including embedded external auditors and coordinated safety standards—that could reshape how frontier AI labs operate. The proposals also test whether voluntary industry coordination can work without government mandates, particularly under an administration that has resisted AI regulation.

Proposed safety mechanisms

Amodei outlined a detailed plan in a recent essay calling for frontier AI companies to grant "ongoing, employee-like access" to independent evaluators. These assessors would receive office space, access badges, and company equipment to monitor safety practices in real time. Anthropic committed to implementing this approach immediately, and Altman said OpenAI would follow suit.

The Anthropic CEO also proposed government-facilitated coordination between leading labs to establish common safety standards and limits on unchecked AI progress. His plan includes multiple levels of international cooperation, from banning obviously dangerous uses like biological weapons development to setting "speed limits" on models capable of recursive self-improvement—systems that can autonomously create improved versions of themselves.

Altman emphasized that pacing does not mean stopping development entirely. "AI progress will continue to be rapid," he wrote on social media. "But it should be slower than it otherwise could be; interventions like safety cases and monitoring have significant costs."

Implementation challenges mount

Several obstacles complicate these proposals. The Trump administration has consistently favored minimal AI regulation to maintain U.S. competitive advantages over China. This geopolitical rivalry creates what experts describe as an "arms race" dynamic that discourages the kind of coordination Amodei envisions.

Domestic competition also poses challenges. Both Anthropic and OpenAI are racing toward potentially record-breaking IPOs, though OpenAI announced it would delay going public this year to focus on safety efforts.

Critics questioned how independent embedded evaluators could truly be and who would set evaluation standards. Aidan Gomez, CEO of Canadian AI lab Cohere, warned that Amodei's proposals could concentrate power further. "If this is the most consequential technology in human history, then the rules for it cannot be written by a small group of commercially aligned companies behind an antitrust waiver," Gomez said.

Sandra Wachter, a professor at the Oxford Internet Institute, called substantial coordination "highly unrealistic" given current global tensions. Elham Tabassi of the Brookings Institution noted that without scientifically valid testing methods, even independent audits would remain "voluntarily-provided, company-controlled" measures.

Path forward uncertain

Nick Reese, former director of emerging technology policy at the Department of Homeland Security, suggested the industry needs to reframe success. "We don't tolerate commercial airline crashes, but we do tolerate significant AI failures," he said.

Whether attention translates to implementation remains unclear. As Zahra Timsah, CEO of governance platform i-GENTIC AI, observed: "Within days, you have CEOs that are competing—all of them, all of a sudden, found common ground. It's receiving attention, but attention is not the same as implementation."

These details were first reported by the Associated Press.