AI Watch, Friday 6 November 2026Global AI stock crash, Wall Street to Asia

The CNIL (France's data protection authority) published a recommendation on 5 November on session replay tools, the systems that record a user's clicks, mouse movements and keystrokes to replay their visit. The regulator targets both the designers of these tools and the sites that install them, and spells out the conditions for compliance, with consent and information first.

The move is a reminder that fine-grained behavioural tracking falls under personal data law, even when it serves to improve an interface.

The French lab announced a new purchase on 5 November, its fifth since January, Maddyness reports. The pace illustrates a strategy of growth through consolidation, as Mistral broadens its offering beyond its language models. The target and the amount were not specified in the source at the time of publication.

The European Commission wants to award a qualification to teenagers who pass an assessment on artificial intelligence, starting in the 2027-2028 school year, Clubic reports. The programme targets three skills: understanding how a model works, checking what it claims, and knowing when to stop it.

Brussels is thus writing AI literacy into the school curriculum, on a par with reading a chart or the digital highway code.

OpenAI unveiled GPT-6 Cyber, a model specialised in detecting software vulnerabilities, verifying whether they can be exploited and proposing fixes, The Verge and ZDNet report. Restricted to certain clients of the "Daybreak Red" programme, it spotted 81 previously unknown flaws in the Firefox browser, which Mozilla has begun to fix.

OpenAI highlights a safeguard: during testing, the model is said to have stepped outside its authorised scope in fewer than 1% of cases, against a far higher rate before. The tool arrives as labs are scrutinised on their ability to secure infrastructure rather than hand attackers new weapons.

Anthropic has opened "Personal Finance", a feature that connects Claude to bank accounts and cards through Plaid, a platform that links up with banks, The Verge and Clubic report. The user can ask "where did my money go this month" and get a view of their spending, a categorisation and answers to their questions.

The feature puts Anthropic on sensitive ground, that of personal financial data, just as the company prepares its stock market debut.

The day after elections in which voters expressed hostility to the expansion of data centres, OpenAI announced new commitments to the communities concerned, including dropping the non-disclosure agreements (NDAs) imposed on local officials, The Verge reports. These clauses had until now kept negotiations on site locations secret.

Meta, for its part, warns that growing political opposition to data centres could complicate its future AI investments. The sector is discovering that its physical expansion now depends on local social consent, not just capital and electricity.

Le Monde reports a lawsuit against Google in the United States: before her parents stepped in, a 13-year-old girl is said to have exchanged thousands of messages with Gemini, the group's AI assistant, on the computer provided by her school, describing her distress and suicidal plans.

The case revives the question of publishers' liability when a minor confides their distress to a conversational agent, and that of the safeguards expected on tools deployed in schools. It adds to a series of lawsuits opened against consumer AI providers over the safety of the youngest users.

China's economic regulator told Japan's SoftBank it would not authorise the $5.4 billion acquisition of the robotics arm of Swiss engineering group ABB, the latter said on Thursday, according to the South China Morning Post. It is the second major deal blocked by Beijing this year, after it ordered Meta to unwind its purchase, for more than $2 billion, of AI start-up Manus.

The decision accompanies a tightening of Chinese control over foreign investment in domestic technology start-ups. For a player that was building its robotics strategy through acquisitions, the signal is clear: robotics has become a strategic asset that Beijing intends to keep on its soil.

Wang Xingxing, founder of Chinese manufacturer Unitree Robotics, said on Thursday at the CIIE trade fair in Shanghai that his robots can now paint and play chess without any prior training, drawing on cutting-edge AI models, the South China Morning Post reports.

He cites OpenAI's GPT-6 model to carry out complex sequences, such as changing the water in a brush or mixing pigments mid-gesture. The claim remains to be verified first-hand, but it sheds light on a dependency that is rarely discussed: Chinese robots driven by an American model, in the middle of a technology export war. It also suggests that value is moving from the robot's body to the model that drives it.