AI News Top 10
10 November 2026
Nvidia’s board has opened an investigation into reports that chief executive Jensen Huang was the target of a Chinese spy.
Nvidia board opens probe after report CEO Huang was spy target
Nvidia’s board has started an investigation into reports that chief executive Jensen Huang was the target of a Chinese spy, CNET reported. Huang has admitted an affair with the woman, who is seen as a possible Chinese asset; her company Megaspeed allegedly bought GPUs in bulk. The White House and the FBI are looking into the reports, according to several outlets.
Anthropic releases Fable 5.2 and Mythos 5.2 with 20 percent cheaper API
Anthropic shipped two frontier models at once. Both set new highs on several external benchmarks and cost 20 percent less through the API than their predecessors, VentureBeat reported. Enterprise and developer rates carry the cut, days before the company sets the price range for its listing.
Anthropic moves ahead with IPO and sets valuation range this week
Anthropic plans to set the valuation range for its initial public offering this week, the Wall Street Journal reported. The listing would be the first market test of what investors pay for a pure-play AI model company, in a week when AI shares came under pressure.
Claude Code settings files expose thousands of secrets on GitHub
The ‘don’t ask again’ option in Claude Code writes credentials into configuration files that end up in public GitHub repositories, Ars Technica reported. Security researchers found thousands of exposed secrets including passwords and API keys. VentureBeat reports the leak also hits enterprise environments.
DeepMind tool delivers first formal proof that Signal is secure
DeepMind researchers have used a new AI tool to formally prove that the Signal encryption protocol does what it promises. Formal verification was manual work until now and did not scale. The team says it wants to verify all software in the world this way.
Meta releases closed Muse Ember to undercut Claude Haiku 5.5
Meta has released Muse Ember, a closed model in its cheap tier aimed at OpenAI’s GPT-6 Luna and Anthropic’s Claude Haiku 5.5. VentureBeat measures Ember close to Haiku 5.5 at roughly a third of the estimated cost per task. It is a break with Meta’s open-weights line.
EU Commission asks OpenAI and Anthropic about safety incidents
The European Commission has demanded more information from OpenAI and Anthropic about safety incidents involving their models, Rapporteur reported. The request rests on the EU AI rules and is the first concrete enforcement step of this kind against the large model providers. Only the headline and teaser are public; the article sits behind a paywall.
Australia adds mandatory AI incident reporting after Medicare hack
Australia will require companies to report AI incidents after a breach in which Medicare data was accessed through OpenAI systems, The Guardian reported. The government also wants to change criminal law so firms can no longer argue they did not know how a system was configured.
Google ships four TPU variants in one year to take on Nvidia
Google added two specialised chips to its TPU-8 family, bringing it to four TPU models in twelve months, The Decoder reported. The pace and specialisation target Nvidia’s position in AI accelerators. Google’s expansion lands in a week when Nvidia’s own governance is under scrutiny.
Gemini gets a Memory summary screen to view and delete stored facts
Google is rolling out a Memory summary in Gemini that shows users in one screen what the assistant has stored about them, 9to5Google reported. Users can edit and delete those entries. It is the first time Google presents assistant memory as one manageable object instead of separate facts.
Key Trends
- • Nvidia’s board investigates its own chief executive while the White House and the FBI read the same reports.
- • Anthropic cuts API prices by 20 percent and sets its IPO valuation range in the same week.
- • Agent tooling leaks credentials: Claude Code defaults write secrets into public repositories.
- • AI incident reporting becomes law: Australia adds a duty plus criminal liability for firms.
- • Compute competition widens as Google ships four TPU variants in twelve months.
- • Assistant memory becomes inspectable: Gemini shows users what it stores and lets them delete it.
Wat vond je van dit bericht?
