AI Watch, Friday 13 November 2026Jensen Huang leaves Nvidia with no named successor

Meeting on 10 November 2026, the European Data Protection Board (EDPB, which brings together the CNIL and its counterparts) chose the use of artificial intelligence in recruitment as the theme of its sixth annual coordinated enforcement action.

Regulators across the twenty-seven member states will jointly investigate the tools that sort applications, score interviews, or shortlist candidates. The EDPB also adopted a statement on South Korea's adequacy and its final recommendations on binding corporate rules (BCRs).

For a French HR department, using automated CV-sorting software now falls within the sights of a concerted European review.

The government will launch ICAID on 20 November, a centre dedicated to artificial intelligence applied to the military, open to allied armed forces and aiming for operational capability during 2027. The project wants to gather foreign partners around a shared technology base, covering intelligence processing, cybersecurity, and decision support. The choice of Rennes draws on a local ecosystem already built around cyberdefence.

Wikipedia has found more than 30,000 references and footnotes drawn from 2,187 sites identified by the outlet Next as generated "in whole or in part", and in some cases "massively", by AI. These sites account for more than 10% of the nearly 20,000 generative-AI sites catalogued in Next's database.

The alliance aims to protect the encyclopedia, whose reliability rests on the quality of its sources, from contamination by automated content that is hard to trace.

Google DeepMind presented Gemini 4 Hydrogen on 12 November, a new model that now powers Google Search's AI mode. The company says it handles complex questions better: multi-part requests, typos that are hard to read, specialised queries, which the model breaks down to process each element separately.

The model is available now in Search, the Gemini app, and every tier of the Google offering. The stakes go beyond raw performance: by placing its latest model inside the engine that billions of people use, Google is trying to hold on to search usage as ChatGPT captures a growing share of queries.

The US carrier United announced that it will not allow AI assistants to buy tickets or manage reservations on its behalf, citing cybersecurity and data protection risks. The decision runs directly against the prevailing talk of agents meant to act for us in buying and planning. It also shows that these agents still generate few real transactions, according to the outlet Next.

The platforms that own the customer relationship set their terms before letting software into the transaction, and security serves here as the first reason to shut the door. TechCrunch,

According to figures reported by Bloomberg and not confirmed by the company, OpenAI passed $10 billion in revenue over a single quarter for the first time, carried by sustained demand for its models. The company nonetheless keeps piling up considerable losses, a sign that its margin remains negative despite surging income.

The same day, Siècle Digital reports that Anthropic is preparing a share sale that would be worth about $250 million to each of its seven founders. The two labs are moving on opposite financial paths: one takes in money and burns it, the other is about to cash in its equity.

China's leading foundry, SMIC (Semiconductor Manufacturing International Corporation, which makes chips on behalf of other companies), expects the breakneck building of data centres to keep AI computing power scarce for two more years, through the second half of 2028.

The forecast lands as the Huang scandal threatens to tighten US restrictions on China's access to advanced processors, which would speed up the shift of Chinese demand towards domestic foundries. For SMIC, every turn of the screw in Washington translates into more orders at home. The Nvidia affair could therefore strengthen the Chinese champion it was meant to contain.

Quanquan Gu, a former AI scientist at ByteDance, has launched a model called NovaAtom-Lite through his new company Geodesic Intelligence, designed to predict the structure and interactions of biomolecules, a key step in the search for new treatments.

According to its founder, the model beats several benchmark systems on the independent FoldBench test suite, across several drug-discovery tasks. The announcement intensifies an already crowded competition around AI applied to the life sciences, where American and Chinese labs meet.

The claimed performance rests for now on the company's own communications, and needs confirmation by third-party teams.