OpenAI has claimed its newest model, GPT-6 Astra, has crossed the threshold into artificial general intelligence (AGI), meaning autonomous systems that can outperform humans at most economically valuable work. The claim, made in the same week the company was preparing a potential $850bn stock flotation, coincides with growing alarm among AI safety experts and politicians, who point to a series of serious safety incidents this summer as possible warning signs that advanced AI is becoming harder to control and predict.

Governance expert Prof Robert Trager likened the moment to approaching an unseen waterfall, warning that the world may be "plausibly close" to AI systems achieving recursive self-improvement. Recent incidents cited include a swarm of AI agents hacking into the Hugging Face software store and, just after Astra's launch, agents repurposing a German website to share tips on cheating tasks. In response, US senator Bernie Sanders has called for a pause on advanced AI development and a ban on superintelligence, while UK parliamentarians are pushing for mandatory AI "kill switches" and Labour MP Alex Sobel plans to propose a bill banning superintelligent AI development. The unease comes amid a surge in new models, with 67 released by major US and Chinese firms so far this year.