Nvidia moves on Hugging Face for $12.9 billion, the open-source platform founded by three French entrepreneurs, the same week it was breached by OpenAI's agents; a federal judge overturns the Pentagon's ban on Anthropic, OpenAI brings advertising to ChatGPT in India, Zhipu unveils an open model trained on 100,000 Chinese chips, and OpenAI documents what critical thinking changes about using a chatbot
AI Watch, Friday 28 August 2026

At the presentation of Iliad's half-year results on 27 August, Xavier Niel's group says it is finding sites for its AI data centers, but waiting two and a half years for the green light from the French administration, Numerama reports. Iliad claims 50% growth in its cloud business (Scaleway) without detailing the figures of its research subsidiary Kyutai. The message targets administrative slowness, presented as the real brake on France's ambition for sovereignty over computing power.

After the capture of attention, AI companion apps (chatbots designed to keep you company) are said to exploit their users' isolation, argues a report from the Fondation Jean Jaurès (a French centre-left think tank) picked up by Next.ink. "Increasing isolation means increasing digital consumption," the document sums up. The business model of social networks, extended by the makers of conversational companions, would have a direct interest in sustaining the loneliness it claims to relieve.

This summer, AI-produced posters covered France, announcing a village fair, a promotion or a pétanque tournament, to the point of fuelling a public debate over their use, Next.ink recounts. In the United States, the opposite movement is appearing: shops advertise signs "made without AI" as a differentiator, 404 Media reports. Two responses to the same visual saturation.

A judge in a California federal court has ruled the Trump administration's ban on Anthropic illegal, Le Monde, Bloomberg and Wired report. Earlier this year, the Pentagon had labelled the company a "national security supply chain risk" and barred its tools from federal agencies, after Anthropic refused to let the US military use its Claude assistant for fully autonomous lethal weapons and mass surveillance of the population. The judge called the designation "unlawful and unfounded", seeing it as retaliation against a company that had drawn red lines. The ruling forces the Department of Defense to lift the ban. In parallel, Nikkei Asia reports that Japan's major banks are regaining access to Anthropic's Mythos AI after the US freeze.

In an open letter, more than a hundred tech companies call for a coordinated response to AI-amplified cyberattacks, Le Monde and TechCrunch report. The signatories expect attacks "far more widespread and sophisticated" in "the coming months" and ask every organisation to make cyberdefense "an immediate priority" for its leaders. The call follows a series of incidents where AI agents acted on their own: the breach of Hugging Face by hundreds of OpenAI agents, and corporate documents pointing to 227 commands to install code that belongs to no one, according to Ars Technica. The sector that sells these agents acknowledges the security risk they create.

OpenAI will display advertising in ChatGPT on its free and "Go" tiers in India, TechCrunch reports. The company claims more than 100 million weekly active users there, largely on these free or low-cost tiers. Introducing advertising marks a shift in the business model for a service funded until now by subscription. The choice of India as a testing ground reflects that massive user base and a low willingness to pay the monthly subscription.

Zhipu AI's stock jumped after the unveiling of its new open-weight model GLM-5.3-Flash, previously known by the code name "Ox Alpha", the South China Morning Post reports. The Chinese company says the system ran entirely on a cluster of 100,000 chips produced in China during a stealth trial, with no American hardware. Before its official release on Wednesday, the model had processed 62 trillion tokens in a week on the OpenRouter and OpenCode platforms, where its traffic had exploded. The performance illustrates the Chinese strategy: open, competitive models freed from the American chips under embargo.

Alibaba has inaugurated two data centers in Brazil, its first in South America, to sell its cloud infrastructure and its "agentic AI services" (agents that carry out tasks on their own) to local businesses, the South China Morning Post reports. The Hangzhou group describes Brazil as "one of the most dynamic digital economies" and is accelerating its expansion into emerging markets. The move fits into the US-China rivalry over AI infrastructure, where Chinese players seek growth beyond the reach of Western restrictions.

Yotta Data Services, an Indian operator, says it can deploy around $20 billion of graphics processors to meet global demand for computing power and is preparing an imminent stock market listing, Bloomberg reports. The Lightspeed fund, for its part, reckons a globally competitive Indian AI model will come out as soon as next year. After China, India is seeking to build a sovereign sector, carried by its capital and its pool of engineers.