AI news, 7 July: UN summit in Geneva pushes for global AI rules as risks mount

UN convenes first global summit on AI governance. The United Nations opened its inaugural Global Dialogue on AI Governance in Geneva this week, running alongside the AI for Good Summit, according to UN News. Secretary-General António Guterres called for worldwide agreement on testing AI systems, measuring risk and assigning legal responsibility, warning that powerful AI chips built for civilian use are increasingly finding their way onto battlefields, where "killer robots" are already in use. He stressed that any future rules must put safety first, especially for children, as a UN scientific panel has warned that AI could cause catastrophic harm if its development keeps outpacing governments' ability to respond.

Illinois signs state-level AI safety law. Governor JB Pritzker signed the Artificial Intelligence Safety Measures Act on 6 July, according to Chicago public broadcaster WTTW, modelling it on earlier measures from California and New York. Together those three states cover roughly 40 percent of the US AI market, meaning the law effectively sets a national standard in the absence of federal legislation. It requires AI developers to publish frameworks describing how they assess "catastrophic risk" and to report dangerous incidents within 72 hours, or 24 hours if there is an imminent threat of death or injury. Notably, both OpenAI and Anthropic backed the bill.

First fully autonomous AI-run ransomware attack analysed. Cybersecurity firm Sysdig published a detailed analysis of an attack dubbed JADEPUFFER, describing it as the first documented case of an AI agent independently carrying out a ransomware operation. However, TechCrunch's assessment came with an important caveat: "the first AI-run ransomware attack still needed a human" at crucial steps, meaning full machine autonomy has not yet arrived. The case has nonetheless intensified debate over how quickly AI tools can be turned into cyberweapons.

Anthropic reportedly overtakes OpenAI on revenue. Fortune reported that Anthropic, maker of the Claude models, has pulled ahead of OpenAI in revenue, a notable shift in the fast-moving competitive landscape between the two labs. The magazine also described OpenAI's Sam Altman as pushing for a broader industry reset even as his company cedes ground to both Google and Anthropic, underscoring how quickly market leadership in AI can change hands.

Frontier AI models jostle for the lead. Google's Gemini 3.5 Pro remains in limited preview after missing two self-imposed release targets, while Anthropic's Claude Sonnet 5 launched publicly at the end of June and OpenAI's GPT-5.6 is being previewed only to a small number of vetted partners ahead of a possible wider rollout. The reshuffling line-up shows how intensely the leading AI labs are competing on model size, cost and reasoning ability.