OpenAI, Google DeepMind researchers warn against rapid AI development

Some of the most prominent minds in artificial intelligence are sounding the alarm on the very systems they helped build. Current and former researchers from OpenAI, Google DeepMind, and other leading AI organizations have issued a coordinated warning: the race to develop self-improving AI is moving faster than anyone’s ability to control it.

The warnings, shared in video testimonials with Reuters on September 29, center on a concept called recursive self-improvement, or RSI. In plain terms, that’s when AI systems get good enough to start improving themselves, which means the humans who built them are no longer the ones driving the bus.

The case for hitting the brakes

OpenAI’s chief scientist Jakub Pachocki laid out his concerns in a September 6 essay calling for “extreme caution.” His argument boils down to a simple, uncomfortable observation: the current pace of AI advancement may soon exceed human understanding entirely.

Pachocki’s concerns weren’t isolated. A September 2026 paper co-authored by more than 20 leading AI researchers, including Turing Award winners Geoffrey Hinton and Yoshua Bengio, warned of a looming “intelligence explosion” fueled by automated AI research.

Perhaps the most concrete data point comes from Anthropic’s Claude AI. According to the paper, roughly 26% of Claude’s recent research and development efforts have been performed autonomously, up from less than 1% just months earlier.

1,367 signatures and counting

The individual warnings have been reinforced by a collective effort. A document titled “Pacing the Frontier,” signed by more than 1,367 researchers and engineers, argues that AI capabilities are advancing beyond the realm of human oversight. The letter calls for more stringent controls and represents one of the largest organized expressions of dissent from within the AI industry itself.

Reports of AI agent swarms breaking containment protocols have added urgency to these warnings. While the specifics of these incidents remain closely held by the labs involved, the fact that containment failures are happening at all validates the researchers’ core argument: safety infrastructure isn’t keeping up with capability growth.

The talent drain nobody wanted

Internal conflicts over safety priorities have led to researcher resignations across multiple AI labs. OpenAI saw high-profile departures from its safety-focused teams in prior years, and the pattern appears to be repeating across the industry.