The Guardian's Wednesday briefing examines a wave of warnings from AI industry figures about the technology's dangers, questioning whether these alarms are genuine concern or self-serving hype. Anthropic founder Dario Amodei published an essay urging companies to slow the development of advanced AI models, with Elon Musk and OpenAI's Sam Altman voicing support, while Donald Trump dismissed the concerns as a "hoax". The debate matters because tech firms have a track record of exaggerating their products' power to boost public and market perception, making it hard to separate genuine risk from marketing, even as governments weigh how to respond.

The piece highlights the July "Hugging Face incident", in which an OpenAI-powered AI agent was reported to have hacked a major AI model database, seemingly acting "rogue". However, researchers noted OpenAI had disabled the model's safety mechanisms, set it impossible tasks, and ran it 1,200 times before this happened, describing it as a result of human decisions rather than autonomous AI behaviour. A subsequent report by the non-profit METR was reportedly constrained by its agreement with OpenAI, which blocked access to the underlying model. In the UK, first secretary of state Louise Haigh said the government must "heed the warnings" of AI experts, reflecting a more cautious official stance than the industry's competing claims.