AI news, 1 September: EU designates ChatGPT under strictest digital rules
EU places ChatGPT under DSA as first major AI service. The European Commission on 31 August designated OpenAI’s ChatGPT a Very Large Online Search Engine under the Digital Services Act, the first time an AI chatbot has faced the bloc’s toughest platform rules, alongside Reddit and Roblox as Very Large Online Platforms. ChatGPT crossed the 45 million monthly EU-user threshold (reporting around 159 million), qualifying as a “hybrid” search service because it answers queries including web search. The companies have four months to assess and mitigate systemic risks around illegal content, minors, mental health, elections and public security, or face fines up to 6% of global revenue. OpenAI said it is preparing to comply. The move gives the Commission stronger investigative powers and marks a milestone in treating advanced AI like major digital platforms, according to the Commission and reports from Euronews and The Verge.
Nvidia invests $3.5 billion in MediaTek for AI chip collaboration. Nvidia announced a $3.5 billion investment in convertible bonds from Taiwan’s MediaTek, its largest direct investment outside the US, as part of a deeper partnership. MediaTek will adopt Nvidia’s NVLink Fusion so customers can design custom AI chips that plug directly into Nvidia rack-scale systems. The deal also covers AI PCs and automotive platforms. Reuters and Bloomberg noted it extends Nvidia’s ecosystem strategy while renewing debate over “circular financing” in the AI hardware sector; Nvidia CEO Jensen Huang rejected that characterisation.
Together AI partners with Saudi Humain on 250 MW data centre. US open-source AI cloud firm Together AI struck a deal with Saudi Arabia’s Humain for 250 megawatts of power and tens of thousands of chips at a new data centre in the kingdom, expecting more than $5 billion in annualised revenue in the first year. The New York Times reported the move reflects growing US community backlash and delays to domestic data-centre projects, pushing companies abroad for capacity. Humain, backed by the Public Investment Fund, aims to position Saudi Arabia as a major AI infrastructure hub.
DeepSeek releases open 305B multimodal model. Chinese lab DeepSeek published DeepSeek-V4-Flash-Vision-Exp, a 305-billion-parameter experimental multimodal model, on Hugging Face under an MIT licence. It adds vision capabilities to the V4-Flash architecture while keeping strong text and agent performance, scoring competitively on multimodal benchmarks. The open weights and matching API pricing lower barriers for developers building vision-enabled agents.
Instagram limits reach of unlabeled AI-generated profiles. Meta’s Instagram renamed its “AI creator” label “AI-generated profile” and will throttle distribution in recommendations and Explore for accounts featuring fully AI-generated personas that fail to disclose it. The change responds to user complaints about deceptive synthetic influencers; ordinary AI editing tools do not require the label. TechCrunch and The Verge covered the transparency push.