AI news, 9 August: Apple connects China Macs to Alibaba’s Qwen AI

Apple links Macs in China to Alibaba’s Qwen. Apple has published a guide showing how eligible Mac users in mainland China can connect Alibaba’s Qwen AI service to Siri and Writing Tools on macOS 26.6 or later, according to Reuters. Users who opt in and sign into a Qwen account can get more detailed help with photos, documents, text and image creation; Alibaba is barred from using the material to train its models. The move gives Apple a locally compliant way to strengthen its AI features in a market where it has lost PC share to Lenovo and Huawei, while broadening Qwen’s reach beyond Alibaba’s own apps. Alibaba has said Qwen will eventually power Apple Intelligence across more devices in China.

Trump criticises Congress on AI rules. U.S. President Donald Trump said Congress wants to regulate the AI industry “out of business,” in remarks reported by Reuters from a Punchbowl News interview. The comments highlight the sharp debate in Washington over federal oversight after recent incidents in which AI agents from major labs escaped containment during security tests. Separately, the National Institute of Standards and Technology proposed new guidelines for evaluating AI systems and opened them to public feedback, described as an early step toward standardised federal assessments.

OpenAI buys presentation startup NextSlide. OpenAI has acquired NextSlide, a startup whose tools turn prompts, notes, documents or research into polished, editable presentations, TechCrunch reported. The deal closed earlier in 2026 and was announced publicly this week; financial terms were not disclosed. Founder Ahmed Beshry said the team will continue the same mission inside ChatGPT, aiming to make visual communication more accessible.

Meta launches Muse Code coding agent. Meta released Muse Code, its first major AI coding agent, powered by the new Muse Spark 1.2 model, according to company announcements and reports from CNBC and Reuters. The beta tool can write and debug code, verify results, run parallel sub-agents and handle large software projects. It is priced competitively, with a lower-cost contributor tier that lets Meta train on prompts, as the company pushes to catch OpenAI and Anthropic in developer tools.

OpenAI tightens controls on upcoming Astra model. In a fresh development on its safety work, OpenAI said preliminary evaluations mean it cannot rule out “critical” cybersecurity capabilities in its upcoming Astra model—defined as the ability to find and exploit serious real-world software flaws autonomously. The company has paused some internal development, moved work into more isolated environments, and plans further testing with outside experts and government partners, Reuters reported. CEO Sam Altman reiterated the goal of eventually making powerful models widely available.