Big Tech's pledge to slow AI development sparks doubts over true motives
Leaders of OpenAI, Anthropic, Google DeepMind and SpaceX loosely agreed over the weekend to slow AI development, but experts are split on whether it's a genuine safety step or a way to dodge real regulation. Critics warn the plan could hurt competitors more than it curbs dangerous technology.
Over the weekend, top AI executives — OpenAI's Sam Altman, Anthropic's Dario Amodei, Google DeepMind cofounder Demis Hassabis, and SpaceX's Elon Musk — loosely agreed to "pace the frontier" of AI development. The proposal calls for embedding independent auditors within labs, domestic oversight of AI companies, and working toward a global agreement on slowing development.
Skeptics, however, argue the pledge may really be aimed at blocking competitors and undercutting open-source AI while avoiding binding legal requirements — some have bluntly called it a "cartel." Industry sources say the reality is more nuanced: the ideas in the proposal have long been championed by AI safety advocates, but many worry it's problematic that industry leaders themselves, rather than independent voices, are the ones steering the conversation.
Tension has grown following a resignation letter from Anthropic researcher Jacob Coxon, who wrote that AI developers privately believe their technology could pose catastrophic risks yet continue racing toward self-improving superintelligence anyway. His letter has been viewed more than 170 million times online.
Many experts point to growing concern over "recursive self-improvement" (RSI) — a hypothetical threshold where AI systems could train and upgrade themselves without human involvement — as the central driver of the current alarm. Anthropic has estimated this milestone could arrive as early as 2027.
A major obstacle to any real slowdown is competition with China, with many in the industry arguing the US cannot afford to fall behind regardless of the risks. At the same time, experts caution that without government regulation, voluntary commitments from AI labs could end up being largely symbolic rather than enforceable.
