OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, Google DeepMind cofounder Demis Hassabis, and SpaceX head Elon Musk have reached a verbal agreement to slow down AI development. The proposal includes embedding third-party auditors, regulating domestic labs, and pursuing a global slowdown. Critics, however, argue this may be a move to stifle competition and avoid meaningful legal oversight.
The discussion follows concerns about recursive self-improvement (RSI), where AI models train and advance without human intervention. Anthropic has suggested this milestone could be reached as early as 2027. Researchers, including those from OpenAI and Anthropic, have publicly expressed concerns that current development speeds pose catastrophic risks, with some estimating a greater than 10 percent chance of human extinction within the next decade.
Industry experts remain divided on the effectiveness of these voluntary commitments. While some see the agreement as a positive step, others warn of 'safety-washing,' where companies perform superficial audits to preempt stricter government regulations. The Tech Oversight Project and other advocates argue that relying on industry self-regulation is insufficient and that government-enforced, measurable standards are necessary.
A significant point of contention remains the role of China, with industry leaders often citing international competition as a reason to maintain development speed. Despite this, some researchers argue that international coordination on AI safety is as critical as nuclear de-proliferation. Meanwhile, the US administration has expressed skepticism toward AI guardrails, with President Trump dismissing recent concerns as a hoax.
