AI news, 6 July: UN opens global talks in Geneva as governments race to write the rules

The most consequential AI story of the past two days is institutional rather than technical. The United Nations' Global Dialogue on AI Governance opened in Geneva today, where member states are beginning discussions on international approaches to managing the technology, according to UN News. The talks draw on a preliminary report from the UN's Independent International Scientific Panel on Artificial Intelligence, which found that AI capabilities are advancing faster than the rules meant to keep them safe, and warned that the window to build effective global governance "remains open but may not stay that way for long." The panel stressed that AI is neither good nor bad in itself; its effects on inequality, democracy and human rights will depend on choices made now by governments and companies.

Security officials are adding urgency to that debate. The Five Eyes intelligence alliance — the United States, United Kingdom, Canada, Australia and New Zealand — issued a joint statement warning that AI-powered cyberattacks capable of automating sophisticated hacking are likely only months away, not years, and urged organisations to prepare defences immediately rather than wait for formal regulation.

Regulation is already reshaping products used by hundreds of millions of people. In China, a new law targeting "AI companion" apps is forcing ByteDance and Alibaba to shut down key features on their popular Doubao and Qwen assistants, according to industry trackers monitoring the rollout. ByteDance's Doubao, which reportedly has around 345 million users, will let people view but not use their saved AI agents and chat histories from mid-July, while Alibaba has offered no way for Qwen users to preserve their agent setups. Both companies apparently judged it easier to switch the features off than to rebuild them to meet the law's anti-addiction requirements.

On the research and product side, Anthropic announced Claude Science, a toolkit for scientists that bundles more than 60 research tools, alongside a new internal program aimed at applying its Claude models to drug discovery for neglected diseases. The announcement follows closely on Anthropic's launch of Claude Sonnet 5 as the default assistant for its free and paid users, illustrating how quickly leading AI labs are pushing from general chatbots toward specialised scientific and consumer tools.

Finally, the business picture is mixed. Venture capital poured into AI at record levels in the first half of the year, with OpenAI and Anthropic alone accounting for roughly 43 percent of a reported $510 billion in global AI investment. At the same time, official US labour data released this week showed only 57,000 jobs added in June, the weakest month since 2024, reviving debate over AI's role in the slowdown. Economists caution the link is not yet proven, but the figures have sharpened scrutiny of automation's effect on hiring.