OpenAI, Anthropic ramp up AI warnings as competition heats up

OpenAI and Anthropic, based in San Francisco, are intensifying warnings about the dangers of advanced AI systems as competition in the sector grows fiercer. Both companies focus on the risk of recursive self-improvement, where AI could autonomously design and train successor models. Anthropic described this scenario as plausible but uncertain last week, while OpenAI predicted it would arrive by March 2028, framing it as both a potential breakthrough and a risk. The companies argue that AI could revolutionize fields like medicine and science but emphasize the need for global guardrails to prevent loss of control. Anthropic co-founder Jack Clark warned that society must retain the ability to 'take its foot off the gas and put on the brake,' while OpenAI has long advocated for an international AI authority similar to the IAEA. Both now propose a multilateral body to slow frontier AI development if necessary, though such coordination faces skepticism. Critics dismiss the warnings as strategic moves. Tech entrepreneur David Sacks and AI researcher Yann LeCun accuse Anthropic of pushing for regulation to stifle competitors, while the 'Stop AI' movement calls the commitments hollow. Gary Marcus, a prominent AI critic, suggested the warnings are pre-IPO marketing to address public backlash. Both companies are racing to raise capital, having filed preliminary IPO applications in early June. Anthropic recently updated its internal guidelines, outlining behaviors the company will never allow and what the industry should avoid—though it won’t enforce the latter unilaterally. The rapid pace of model releases, roughly every four to six weeks, underscores the urgency to stay ahead in the AI arms race.