Uncategorized

DailyAI Briefing: Jensen Huang resigns as Nvidia CEO, plus 4 more

Welcome to the DailyAI Briefing, the day’s five biggest AI stories. Jensen Huang has resigned as chief executive of Nvidia, and its shares ended the day far above their early low. Google made its new Gemini 4 Hydrogen model the default for AI answers in Search, OpenAI launched ChatGPT Wallet so that its AI agents can pay for shopping, experts advising the French president called for safety tests before chatbots launch, and a new survey found that teens in richer families now use AI for homework more than teens in poorer ones.

Let’s dig in.

Jensen Huang resigns as head of Nvidia

4.3%. That is how much Nvidia shares fell on Thursday, the day the company said Jensen Huang had resigned as chief executive. Early in the session the drop was about 14%. Those figures were reported by a news agency.

Huang co-founded the company and ran it for 33 years. A filing with the market regulator shows that the board accepted his resignation on Wednesday. Colette Kress, the chief financial officer since 2013, is now interim chief executive and keeps her finance job. Harvey Jones, a board member since 1993, becomes the first chair of the board.

The company gave no reason. The resignation came after a week of reports about his private life. Our report explains who runs the company now. The real test comes on Tuesday 17 November, when the chipmaker publishes its third-quarter results with Kress in charge. So many AI models are trained on its chips that those numbers will say more about AI spending than any change at the top.

Google puts Gemini 4 Hydrogen behind its AI search answers

2.5 times faster. That is Google’s own figure for how quickly Gemini 4 Hydrogen finishes tasks compared with Gemini 3.8 Flash, the model it replaces. The new model was announced in a company blog post on 12 November. It is now the default model in AI Mode and handles the hardest questions in AI Overviews, the summaries at the top of search results.

Google says the model scores 57% on Terminal-Bench 4.1, a test of coding tasks done in a computer terminal, against 18.2% for the older model. Nobody outside Google has checked these figures yet. Our article explains what changes for people who search. Most people will meet this model without ever choosing it, so independent tests of its answers matter more than the speed claim.

ChatGPT Wallet lets AI agents pay for shopping

1.2 billion. That is how many people OpenAI says use ChatGPT each week, but only a small group of them can try ChatGPT Wallet. It went live on Friday for paying Pro users in the United States who are 18 or older. The new payment feature was announced in a company blog post on 13 November.

OpenAI built it with American Express, Mastercard and Visa. When the AI agent is ready to buy something, the user sees the items, the total and the card, then approves the purchase with Face ID or a fingerprint scan. A stand-in card number is used in place of the real one, so OpenAI says neither the shop nor the AI agent sees the real card details.

A safe payment is not the same as a safe shop, though. In September, a firm that sells brand protection ran its own tests and found that ChatGPT pointed to a risky website, such as a fake shop, in 8% of its shopping answers. Our report looks at what the new feature does and does not protect. For now, the best safeguard is still a person who reads each purchase before approving it.

French experts want chatbots tested before launch

18 recommendations. That is what a panel of nine experts set up by the French presidency handed to Emmanuel Macron on 10 November. The presidency has published a summary of the report on its website. Chatbots for the general public, including those built into social networks, would have to show that they do not exploit or worsen the weaknesses of the people who use them, with a second check after every major change to the model.

The panel also wants limits on chatbots that act too much like a human or flatter their users, a ban on hidden advertising in chatbot answers, and approval and age limits for toys with a chatbot inside.

None of this is law yet, and France cannot bring in such checks on its own. Our article sets out the main recommendations. A test before launch would reverse the usual order, in which rules for AI arrive only after the harm.

Teens in richer families use AI for homework more

57% against 43%. Those are the shares of teens in the United States who use AI for homework in families earning $150,000 or more a year and in families earning less than $50,000. A year earlier, both groups were level at 39%. Those numbers come from a new report published on Thursday.

RAND surveyed about 1,000 parents and 1,800 teens aged 12 to 18. Teens tend to copy what their parents do, and 67% of parents in the highest-income group encourage their teen to use AI for school, against 45% of the lowest-income parents.

RAND does not say whether the new difference helps or hurts richer teens. Research cited in the report finds that learning suffers when AI replaces a student’s own first try at a task, but not when AI is used to improve work the student has already done. Our article covers the full survey and its advice for parents. How teens use AI at home may matter as much as any school rule.

Friday’s common thread

A chief executive stepped down, a search engine changed its default model and a chatbot started paying for shopping. The other two stories were about who sets the rules for AI, whether that is a panel of experts in Paris or a parent at the kitchen table.

See you tomorrow.

The post DailyAI Briefing: Jensen Huang resigns as Nvidia CEO, plus 4 more appeared first on DailyAI.

ChatGPT Wallet Lets OpenAI’s Agents Pay Without Seeing Your Card

OpenAI’s AI agents can now pay for shopping on a user’s behalf without seeing the real card number. The new ChatGPT Wallet went live on Friday, but only for paying Pro users in the US, a small group among the 1.2 billion people OpenAI says use ChatGPT each week.

The new payment feature was announced in a company blog post on 13 November. OpenAI built it together with American Express, Mastercard and Visa. For now it works only with dots, the always-on agents that OpenAI added to ChatGPT in September. Users must be 18 or older and have an eligible card issued in the US.

How a purchase works

A user adds a card once. When the AI agent is ready to buy something, the user sees the items, the total and the chosen card. The user then approves the purchase with a check such as Face ID or a fingerprint scan.

After that, the card network creates a payment token. This is a stand-in card number, security code and expiry date that is used in place of the real card details. OpenAI says neither the shop nor the AI agent sees the real card. Users keep their card rewards and the usual protections, for example when an item arrives broken.

Each agent has its own cloud computer and web browser, and it can keep working on a task after the user closes the chat. The agents were first shown at a developer event on 29 September.

Plans for purchases without approval

OpenAI says that over time, users should be able to let AI agents make purchases without approving each one. Limits such as a spending cap, a type of product or a time period would apply. It has not given a date.

Some shops may also turn agents away by mistake. OpenAI says some websites and fraud systems block or cancel real purchases made by agents. It is working with payment firms including Adyen, Checkout.com and Stripe to reduce this.

Luke Gebb, who leads innovation at American Express, said the limited start was the “right approach to make sure things are working”. He made the comment in an interview with an American business news channel.

Paying safely is not the same as buying safely

A safe payment does not mean a safe shop. In September, a firm that protects brands asked ChatGPT, Gemini and Perplexity where to buy products from 151 brands. Across 6,795 answers, ChatGPT showed a risky website, such as a fake shop or a dead web address, in 8% of its answers. Perplexity did so in 15% and Gemini in 4%.

The firm has published its method and full results. It sells brand protection, so these are its own figures from its own test questions. Not every business wants AI agents to shop on behalf of its customers, either. We reported this week on one airline that blocks them.

Why it matters: The token keeps the card number away from the AI agent, but it does not check whether the shop is real. For now, the main safeguard is still the user, who must read each purchase before approving it.

The post ChatGPT Wallet Lets OpenAI’s Agents Pay Without Seeing Your Card appeared first on DailyAI.

Google’s Gemini 4 Hydrogen Is 2.5 Times Faster and Now Powers Search

Google says its new Gemini 4 Hydrogen model finishes tasks 2.5 times faster than the model it replaces. Since Thursday it has been the main model behind AI answers in Google Search, so most people who search will meet it without choosing it.

The new model was announced in a company blog post on 12 November. Gemini 4 Hydrogen is the fast, cheaper member of the Gemini 4 family. It sits below Gemini 4 Carbon and Gemini 4 Argon, the larger models Google aims at the hardest problems. It takes over from Gemini 3.8 Flash, which came out in September.

What changes for people who use Google

In Search, Gemini 4 Hydrogen is now the default model for AI Mode, the chat-style search page. Google also uses it for the hardest questions in AI Overviews, the summaries at the top of results. The company says it copes better with long questions, typos and questions in several parts, and that it shows a wider range of links to websites.

In the Gemini app, paying users can choose the new model from a list of models. Free users get the new model only when the app’s automatic setting picks it for a question.

The numbers Google gives

All of these figures come from Google and have not been checked by anyone else. Google says the model scores 57% on Terminal-Bench 4.1, a test of coding tasks done in a computer terminal, against 18.2% for the older model. It also says the new model makes up facts more than five times less often.

Artificial Analysis, an independent group, runs a ranking of AI models called the Intelligence Index. Google says the new model scores 51 on this index at its default setting. That is one point more than Gemini 4 Argon, the larger model. At its highest reasoning setting, the score rises to 54.

For developers, it costs $1.50 per million tokens of input and $7.50 per million tokens of output. A token is a small piece of text, often part of a word. There is also a free tier for developers, with limits on use.

What Google says about safety

Google says the new model is up to two times less likely to get around permission limits when a task is blocked. The model card lists known limits and safety test results. Like the speed figures, these are Google’s own test results.

The launch comes as AI companies compete on price and speed. Anthropic made its Claude Fable 5.2 model 20% cheaper this week. We reported on that price cut on Tuesday. Earlier this month, OpenAI said ChatGPT now runs about a fifth as many searches as Google. We covered those figures in a separate article.

Why it matters: Search is where most people meet AI without asking for it. A change to the model behind AI Mode changes the answers that a very large number of people see. So far, Google has not published independent checks of its claims.

The post Google’s Gemini 4 Hydrogen Is 2.5 Times Faster and Now Powers Search appeared first on DailyAI.

French Experts Want Consumer Chatbots Tested Before Launch

A panel of nine experts set up by the French presidency wants consumer chatbots to pass a safety test before they can be sold in Europe. Its report was handed to Emmanuel Macron on 10 November and makes 18 recommendations. None of them is law yet.

The presidency has published a summary of the report on its website. The group was led by the AI researcher Serena Villata and two psychiatrists, Raphaël Gaillard and Amine Benyamina. It held 55 hearings with doctors, researchers, companies and child protection groups.

What the experts want

The main idea is a check before launch. Chatbots for the general public, including those built into social networks, would have to show that they do not exploit or worsen the weaknesses of the people who use them. A second check would follow every major change to the model, or a warning from users.

The panel also wants strict limits on design tricks that are meant to keep people using a chatbot for longer. It names two of them. One is a chatbot that acts too much like a human. The other is sycophancy, which means a chatbot that flatters users and agrees with whatever they say.

Toys with a chatbot inside would need approval before they could be sold, as well as age limits. Hidden advertising in chatbot answers would be banned, and companies would have to explain how they make money from their chatbots.

The numbers behind the report

The summary says 33 million people in France used generative AI at the start of this year. A quarter of French people say they worry about its risks.

Among French teenagers who use these tools, 92% say they spend more time than they would like on at least one AI app. The figure comes from a survey by the think tank Interface with YouGov. For AI companion apps, which are designed to act like a friend, the share is 59%. These figures are set out in the full report. The panel adds that heavy use of an app is not, on its own, a medical sign of addiction.

Next steps for the report

France cannot bring in these checks on its own. The report says several of its ideas could shape the French position in talks on EU rules, including an update to the EU AI Act and a proposed EU law to protect children online.

On Thursday, four AI companies agreed to independent child safety tests under a charter presented in Paris. We covered that agreement in a separate article. At the time, Macron said no AI should reach children before it has been tested. The new report explains how a rule like that could work. Australia is also taking steps against chatbots that act as children’s best friends.

Why it matters: Most rules for AI apply after harm has been done. A test before launch would make companies prove that a chatbot is safe for the public before anyone uses it.

The post French Experts Want Consumer Chatbots Tested Before Launch appeared first on DailyAI.

Nvidia Shares Fall 4.3% After Jensen Huang Resigns as CEO

Nvidia shares fell by about 14% when trading opened in New York on Thursday, after the company said Jensen Huang had resigned as chief executive. By the end of the day, most of that loss had gone. The shares closed 4.3% lower, and the company was still worth around $4 trillion.

Those figures were reported by the Associated Press. In plain terms, many investors sold in a panic at first, then decided that the business had not changed overnight. Even so, the share price is now 30.5% below the record it set on 6 October.

Who runs Nvidia now

Huang co-founded Nvidia and led it for 33 years. He has left his jobs as president, chief executive and board member with immediate effect. A filing with the market regulator shows that the board accepted his resignation on Wednesday.

Colette Kress, 59, has been the chief financial officer since 2013. She now becomes interim chief executive and keeps her finance job at the same time. The board will consider internal and external candidates for the permanent job. Harvey Jones, 73, who has sat on the board since 1993, becomes its first chair.

“The Board has full confidence in Colette and the leadership team, and we expect the company’s operations, customer commitments and product roadmap to continue without interruption,” Jones said. His comments were published in the company’s announcement.

Why he left

The company gave no reason. The resignation followed a week of reports about Huang’s private life. On Sunday, he told staff that he had been unfaithful to his wife. Earlier reports said US officials believed the woman involved was working for Chinese intelligence. China has called those claims groundless. We reported on Monday that the board had opened an investigation.

On Wednesday, the New York Times reported a second affair, this time with a senior employee who reports to him. His resignation was made public the following morning.

The next test is on Tuesday

Nvidia sells the chips that train and run many of the largest AI models, so its results are watched as a guide to spending across the whole AI industry. The company says it will publish its third-quarter results on Tuesday 17 November, as planned. Its call with investors starts at 2pm Pacific time, which is 23:00 in Paris. It will be the first results day with Kress in charge.

Other AI companies will be watching too. Anthropic rents large numbers of these chips through cloud providers. We reported this week on its plan to raise up to $64bn by selling shares.

Why it matters: Nvidia’s share price is one of the main measures of the AI boom. Tuesday’s numbers will show whether its customers are still buying chips at the same pace now that its founder has stepped down.

The post Nvidia Shares Fall 4.3% After Jensen Huang Resigns as CEO appeared first on DailyAI.

Teens in Richer Families Now Use AI for Homework More, RAND Finds

In the United States, 57% of teens from families earning $150,000 or more a year now use AI to help with their homework. In families earning less than $50,000, the figure is 43%. Only a year earlier, both groups were level at 39%.

Those numbers come from a new report published on Thursday. RAND, a research group in the United States, surveyed about 1,000 parents and 1,800 teens aged 12 to 18. The surveys were run in spring 2025 and again in May 2026. The Gates Foundation paid for the work.

The report’s main finding is simple. Teens tend to copy what their parents do. Parents with higher incomes have taken up AI fastest, and they are also the most likely to tell their children to use it for school. As a result, a difference in homework habits has opened up between richer and poorer homes. RAND puts it at 13 percentage points, after rounding.

Parents took up AI first

Among all parents in the survey, the share who used AI for work or daily life rose from 54% in February 2025 to 72% in May 2026. Most of that rise came from frequent users, meaning people who use AI at least once a week.

Income made a large difference. In the highest-income families, 84% of parents now use AI. In the lowest-income families, 63% do. The difference between the two groups grew from 15 to 21 percentage points in 15 months. RAND believes higher-income parents use AI more because better-paid office jobs use it more often.

What parents tell their children

RAND asked parents which statement best described what they tell their child about AI and schoolwork. Overall, 56% said they encourage it to some degree. Another 24% said they tell their child not to use it at all for school, and 20% said they had not talked about it.

Again, income mattered. About two-thirds of parents in the highest-income group, 67%, encourage their teen to use AI. Among the lowest-income parents, 45% do. Most parents who encourage it set limits, for example using AI to brainstorm ideas or to check finished work. However, 17% of the highest-income parents said they encourage their child to use AI as much as school rules permit.

A parent’s own habits made a difference too. Among parents who use AI often, 22% encourage their teen to use it wherever possible. Among parents who never use it, almost none do. One result surprised the researchers. About half of parents who agree that AI harms critical thinking, 53%, still encourage their teen to use it for school.

Teens follow their parents

To test the link directly, RAND matched 533 teens with 427 parents from the same homes. Among teens whose parents use AI often, 63% use it for homework in at least some classes. Among teens whose parents never use it, only 40% do.

What teens believe their parents think mattered even more. Of teens who said their parents encourage AI for school, 81% use it for homework. Of teens who said their parents discourage it, 38% still do. RAND found this was linked to AI use more strongly than any other factor it tested, including family income and sex. Only age mattered more.

The matched group is small and was not designed to represent all US families, so RAND advises some caution with these figures. The national sample of 1,807 teens shows a similar link. Of teens with no one encouraging them, 33% used AI for homework. When a parent was among those encouraging them, the figure was between 55% and 77%.

The report also shows that teens hear mixed messages. In May 2026, 78% of teens said at least one type of person had told them not to use AI for school. Teachers were the people most often named as telling them not to, at 57%. Friends were the most common source of encouragement, at 36%.

Parents and teens also do not agree on what was said at home. Among teens whose parents said they encourage AI use wherever possible, only 25% said their parents encouraged them. Another 27% believed their parents were against it.

Why the first attempt matters

RAND’s concern is not that teens use AI at all. It is about when they use it. The report cites research that includes four randomised trials. It finds that learning suffers when AI replaces a student’s own first try at a task, such as writing an essay or solving a maths problem. Using AI to improve work the student has already done does not show the same harm.

One of those trials gave nearly 1,000 high school students in Turkey access to GPT-4 during maths practice. Students who used a plain chatbot scored 48% higher on practice problems. In the exam that followed, taken without AI, they scored 17% lower than students who never had it. A version that gave hints instead of answers removed most of that harm.

This is why RAND does not say whether the new income difference helps or hurts richer teens. If they mostly let AI do the work, their results could fall, which would shrink the long-standing achievement gap between richer and poorer students. If they use AI to check their work, or to plan for college and careers, the gap could grow instead.

Heather Schwartz, the report’s lead author and a vice president at RAND, put it plainly. “It’s not simply a question of whether students should use AI, but how they use it,” she said. Her comments were published with the report. “Teens need guidance on when AI can support learning and when it risks doing the thinking for them,” Schwartz said.

How this fits with other surveys

Other surveys also show that AI is now a normal part of schoolwork. In February, the Pew Research Center reported that 54% of teens in the United States aged 13 to 17 had used a chatbot for help with schoolwork. One in ten said they do all or most of their schoolwork with a chatbot’s help.

Income did not seem to matter much before this. In autumn 2024, Pew asked teens whether they had used ChatGPT for schoolwork. Its survey found no difference between richer and poorer homes. That survey asked about one product, so it is not a direct comparison. Still, RAND’s own data shows that teens in the two income groups used AI for homework at the same rate in spring 2025, and at different rates a year later.

Teens use AI for more than homework. Last month we reported that most teens in a large student survey turn to AI when they feel low. Schools are also having to decide how much access to allow. We have also argued that schools should not be forced to pick between AI tools and web search.

What RAND wants schools, parents and companies to do

  • Schools and districts: send families clear, evidence-based talking points on why using AI without limits on a first attempt can harm learning, and use settings on school devices to block general chatbots while students make that first attempt.
  • Parents: talk openly with teens about the difference between AI doing their work and AI helping them improve it, and look at settings on phones and tablets that limit open-ended AI use.
  • AI companies: build features that tell helpful use apart from handing over the thinking, rather than blocking students outright.

Why it matters: Homework is where most students practise on their own, and AI use there is now split by family income. Whether that helps or harms richer teens depends on how they use it, which is why RAND says parents need to be far more explicit about the rules at home.

The post Teens in Richer Families Now Use AI for Homework More, RAND Finds appeared first on DailyAI.