AI news, 9 July: OpenAI's newest models go fully public after a government review, as the Geneva debate over AI safety heats up

OpenAI's GPT-5.6 goes public after a government check. OpenAI said its newest model family, GPT-5.6 (branded Sol, Terra and Luna), will roll out to everyone this Thursday, according to Engadget and Axios. The models had been limited to a small group of government-approved users since late June, after the Trump administration asked OpenAI to submit them for extra testing under a June executive order on AI cybersecurity. A White House official told Axios that no formal permission was required, saying decisions on release timing "rest entirely with the companies," but the episode has fuelled debate over how much informal influence governments now have over AI releases.

OpenAI gives ChatGPT a more natural voice. OpenAI also introduced GPT-Live, a new voice system now powering ChatGPT Voice. Unlike earlier versions that processed speech in separate steps, GPT-Live uses a "full-duplex architecture, meaning it can listen and speak at the same time," OpenAI said, letting it interject naturally or pause when a user needs a moment. For harder questions, it quietly hands off to a more powerful reasoning model behind the scenes before responding, and OpenAI says it added extra safeguards for teenage users and conversations involving self-harm.

World leaders and AI CEOs meet in Geneva over safety fears. A newly formed UN AI for Good Global Commission held its first working meeting this week, bringing together heads of state with the chief executives of Nvidia, Amazon and Microsoft, according to reporting on the summit. The meeting followed the UN's Global Dialogue on AI Governance in Geneva, where an independent scientific panel co-chaired by Yoshua Bengio warned, per UN News, that researchers cannot currently rule out AI causing severe harm as the technology advances faster than oversight can keep up. Journalist Maria Ressa separately raised alarms about AI-driven misinformation eroding trust in democracies.

Chinese AI models gain ground with businesses. A CNBC investigation found that Chinese-made AI models now account for between 30% and 46% of enterprise token usage on major US developer platforms, up from roughly 11% the prior year, a shift largely attributed to Chinese models being far cheaper than leading US alternatives. Separately, industry reporting this week described Anthropic as having overtaken OpenAI in annual revenue for the first time, underscoring how quickly the competitive order among AI companies is shifting.

Illinois signs a tough new AI safety law. Governor JB Pritzker signed the Artificial Intelligence Safety Measures Act, modelled on earlier California and New York laws that, lawmakers estimate, together cover roughly 40% of the US AI market, according to Illinois public broadcaster WTTW. The law requires AI developers to publish a framework for assessing "catastrophic risk" and to report safety incidents within 72 hours, or 24 hours if there is imminent danger of death or serious injury, with lawmakers citing earlier AI-linked security and violence incidents as justification.