Tech leaders split on slowing frontier AI to improve safety

In short: More tech leaders are warning that top AI systems are improving faster than safety checks, and they disagree on whether to slow down and add shared rules.

What's going on

A growing number of executives working on frontier AI, meaning the most advanced AI models, say progress is moving faster than safety systems can keep up. They are calling for slower development, common standards across companies, and stronger independent oversight. Independent oversight means outside experts get to test and review systems, instead of companies grading their own homework.

One clear split is speed versus safeguards. Anthropic CEO Dario Amodei has argued for an intentional slowdown so testing and protections can catch up. His proposals include stronger testing, independent evaluations, and coordination among leading AI companies, with the goal of international cooperation later.

Other leaders have taken different positions. Reuters and NPR have reported that OpenAI CEO Sam Altman and some others have shown support for coordination or pacing. At the same time, leaders such as Meta CEO Mark Zuckerberg and Nvidia CEO Jensen Huang have pushed back on new collective restraints or added regulation.

The debate is also becoming more concrete. Reuters reported discussions among Anthropic, OpenAI, and Google about creating a new AI safety body. It could work a bit like FINRA, a self-regulatory group in finance (like an industry rulebook with enforcement, created by the industry, not the government). Reuters also reported that Amodei’s approach starts with better internal safety work, then shared standards across major labs, and then a global system.

What to watch

Watch whether big AI companies announce a shared safety group or shared testing standards, and whether governments treat these efforts as enough or decide to write stronger rules of their own.

Source: NYTimes