OpenAI predicts recursive AI by 2028; Anthropic warns of loss of control

OpenAI and Anthropic, the makers of ChatGPT and Claude, have sharpened their warnings about recursive self-improvement—a scenario where AI systems design and train the next generation of AI with minimal human involvement. OpenAI went further than Anthropic by predicting this milestone would arrive by March 2028, framing it as an automated researcher that could speed up scientific discovery and deliver a personal AI for everyone. Anthropic, while acknowledging potential breakthroughs in medicine and technology, has openly warned of “loss of control,” urging the industry to have the option to “take your foot off the gas.” The companies release new models every four to six weeks, showing the accelerating pace of AI development.

To address these risks, both firms have floated ideas for international coordination. Anthropic proposed a global system to temporarily suspend frontier AI development—but only if no competitor takes advantage of the pause. OpenAI suggested a multilateral body akin to the IAEA to limit AI capability growth. However, critics including former Trump AI adviser David Sacks accuse the companies of “fear-mongering” to sway investors. The warnings come as both OpenAI and Anthropic have filed preliminary IPO applications in early June, signaling a rush to raise capital while simultaneously pushing for industry-wide guardrails. The credibility of their caution is questioned, as their financial ambitions and product release cadence seem to contradict the calls for a slowdown.