Uncategorized

DailyAI Briefing: Gemini may breach press rights, plus 4 more

Welcome to the DailyAI Briefing, the day’s five biggest AI stories. A top adviser to the EU’s highest court says chatbots like Gemini can break publishers’ rights when they reuse news articles, a day after the New York Times sued Brave for selling its journalism to AI companies. OpenAI promised that families won’t pay for its electricity, it is also offering up to $172,000 for the worst ChatGPT bugs, and new data suggests people who use AI at work are putting in longer days, not shorter ones.

Let’s dig in.

Gemini may breach press rights, says EU court adviser

In the first generative AI case to reach the Court of Justice of the European Union, Advocate General Maciej Szpunar said on Tuesday that a chatbot which fetches a news article, reuses it in an answer and hands that answer to a user “may, in certain circumstances” infringe the publisher’s rights. The case, C-250/25, was brought by Hungarian web publisher Like Company against Google over Gemini answers given between June 2023 and May 2024, according to the court’s press release.

He drew two lines. A chatbot is on the wrong side if it gets at the article unlawfully, for example by getting round a paywall or anti-scraping measures, or if its answer reuses the article’s wording so recognisably that the reader no longer needs the original. Repeating facts is fine, since facts aren’t protected. On training, he called the questions hypothetical but said using protected material to train a model is “as a rule” allowed under EU law’s text and data mining exception. The opinion isn’t binding, though the court usually follows its advisers. The plain reading: in Europe, the fight is moving from what models learn to what chatbots repeat.

The New York Times sues Brave over AI scraping

More than 30 billion web pages. That is the size of the search index Brave built with a crawler publishers couldn’t identify or block, according to the New York Times’ 72-page complaint, filed in Manhattan federal court on Monday. The Times says Brave sells that content through its search API to AI companies it names as Anthropic, Perplexity, Mistral, xAI and Cohere, none of which are defendants, and that Brave’s browser has a paywall blocker. The New York Daily News, Chicago Tribune and Orlando Sentinel filed a similar suit the same day, TechCrunch reported.

The Times says it sent Brave’s chief executive Brendan Eich cease-and-desist letters in August 2025 and April 2026 and got no reply. Brave didn’t respond to requests for comment from The Verge, while Perplexity said the Times “is overpaying its attorneys and underpaying its journalists”. Read alongside the EU opinion, the same issue keeps coming up on both sides of the Atlantic: getting past a paywall is what turns reading the news into taking it.

OpenAI says it will pay its own power bill

Up to 1,000 megawatts. That’s how much demand OpenAI’s planned Georgia campus can give back to the grid when it is under strain, under a 25-year contract with Georgia Power, out of an expected load of about 3,200 megawatts. OpenAI has now made that deal the model for a national scheme it calls Grid Protectors, and says “families and small businesses should not be asked to subsidize a compute campus’s electricity needs”.

There’s no list of new sites, megawatts or dates yet, and the Georgia contract hasn’t been published. The number households will care about comes from the utility, not OpenAI: Georgia Power says its big-customer contracts should save a typical home at least $15 a month from 2029. That’s a forecast, and so far the whole programme rests on one deal. Our deep dive goes through what it means for your bill.

OpenAI will pay up to $172,000 for the worst ChatGPT bugs

A one-click ChatGPT account takeover that used to earn a researcher $6,500 could now pay $51,577 with every bonus stacked, under the bigger bug bounty rewards OpenAI announced on Monday. The top payout for the rarest finds, such as reading any user’s chats, rises to about $172,000. OpenAI is also bringing in Cathedral Systems for its next internal security sprint.

Cathedral is the start-up whose AI agents got into parts of Anthropic’s internal systems last month, The Washington Post reported, an episode OpenAI’s posts don’t mention. Bigger rewards make it likelier the next hole in ChatGPT gets reported rather than sold, which matters if your chats hold anything personal. Our story has the full rate card.

AI users are working longer days

People who use AI at work are active for up to 90 minutes longer a day than colleagues who don’t, according to ActivTrak‘s screen-activity data on 9,140 workers at 467 companies. Heavy users also had shorter stretches of focus, 37.8 minutes against 48.2 for light users, and those using a chatbot in a separate browser tab switched apps up to 4.2 times an hour.

ActivTrak sells workplace monitoring software, and its data shows a link, not a cause: people who already work long days may simply be keener on AI. Still, if the tools make tasks faster and the day gets longer anyway, the time saved is going to employers rather than workers. We dug into the numbers this afternoon.

Signing off

Today was about who pays. Publishers want chatbots and search companies to pay for their journalism, OpenAI says it will cover its own electricity and bigger bug bills, and workers may be paying for AI with their evenings. Earlier, we also covered UK regulators telling firms they can’t blame their AI agents when things go wrong.

See you tomorrow.

The post DailyAI Briefing: Gemini may breach press rights, plus 4 more appeared first on DailyAI.

AI Was Meant to Save Time. Its Users Are Working Longer Days

People who use AI at work are online for up to 90 minutes longer each day than colleagues who don’t, according to activity data from 9,140 office workers. The tools are sold as a way to save time. In this data, the time isn’t coming back.

The figures come from a report published on 29 October by the ActivTrak Productivity Lab. ActivTrak sells software that tracks how staff spend their time on work computers, and the lab drew on that data from 467 companies between 1 April and 25 September 2026. It measured what people actually did on screen rather than asking them. Gizmodo picked up the findings on Tuesday.

What the data shows

A “workday span” here means the time from someone’s first activity of the day to their last. For AI users it averaged 8.4 to 9.1 hours, depending on how much they used the tools. For people who don’t use AI it was 7.6 hours.

Heavy users, defined as more than 45 minutes of AI a day across at least five sessions, spent less of their time in meetings: 3%, against 5% for non-users. Their longer days still added about 18 minutes of work outside scheduled hours.

Their attention was also more broken up. Heavy users averaged 37.8 minutes of uninterrupted work at a stretch, against 48.2 minutes for light users, about 22% less. People using a chatbot in a separate browser tab switched between apps up to 4.2 times an hour, more than four times the non-user rate of 0.95. Those using AI built into their main work software switched about 0.6 times an hour.

What it doesn’t show

The report shows a link, not a cause. People who already work long days may simply be more likely to take up AI tools. ActivTrak’s press release goes further, quoting Gabriela Mauch, head of the Productivity Lab: “AI may help employees complete tasks faster, but our research shows it contributes to longer workdays, more application switching and shorter focus sessions.”

It fits earlier research. In an eight-month study at a US tech company of about 200 people, published in Harvard Business Review in February, UC Berkeley Haas researchers Aruna Ranganathan and Xingqi Maggie Ye found that employees “worked at a faster pace, took on a broader scope of tasks, and extended work into more hours of the day, often without being asked to do so.”

Why it matters

If AI makes tasks quicker but the day gets longer, the saving is going to employers as extra output, not to workers as free time. For anyone told AI will lighten their load, the early evidence says to watch the clock, and to ask what happens to the hours it saves.

The post AI Was Meant to Save Time. Its Users Are Working Longer Days appeared first on DailyAI.

OpenAI Will Now Pay Up to $172,000 for the Worst ChatGPT Bugs

$171,925. That is what OpenAI now says it could pay one outside researcher for a flaw that lets someone read any ChatGPT user’s chats. The company raised its bug bounty rewards on Monday and said it is bringing in Cathedral, the start-up that got inside parts of Anthropic’s systems last month, to attack its own defences.

A bug bounty is a standing offer of cash to people who find security holes and report them privately instead of exploiting them. In its announcement, OpenAI said it is “more than quadrupling” the midpoint of its reward ranges for medium, high and critical findings, on average. A critical bug in ChatGPT, Codex or its developer platform now pays up to $30,000. The rarest, most serious finds can still earn up to $100,000 before bonuses, and up to $172,000 with them.

What the new rates look like

OpenAI gave examples, which it calls illustrative. A one-click ChatGPT account takeover that once earned $6,500 would now pay $51,577 with every bonus stacked. A medium bug that let a read-only team member upload files to a Project goes from $750 to $2,415.

The bonuses come from three new multipliers: one for high-quality reports with a working proof, a five-tier loyalty scheme for repeat researchers, and short campaigns with extra money for newly launched features. Reports still waiting for payment get the new rates, and anyone paid in the previous 30 days gets the difference. The programme runs on Bugcrowd.

Hiring the firm that got into Anthropic

In a separate post, OpenAI said Cathedral’s researchers will use OpenAI’s models alongside its own teams in its next internal security sprint. Cathedral chief executive Gavin Kliger said OpenAI was “setting an important example” by working with outside firms.

Cathedral, founded by four former DOGE staffers, said its AI agents found paths into Anthropic’s Slack messages, meeting transcripts and salary records in under 12 hours, The Washington Post reported on 26 October. Anthropic said only “a limited set” of its sales, HR and IT systems were reached, and that Claude chats, source code and model weights were not. It also said Cathedral didn’t disclose all of its activity under the bounty rules, Decrypt reported. OpenAI’s posts don’t mention any of this.

In its last sprint, OpenAI said, it put 25% of its product engineers on finding and fixing weaknesses, using its Codex coding agent. Neither company has said when the new one starts or what Cathedral will be paid.

Why it matters

Bigger rewards make it more likely that the next hole in ChatGPT is reported to OpenAI rather than sold or used. That matters to anyone who keeps personal or work material in their chats, which can leak through bugs like the shared-link flaw we reported on Monday.

The post OpenAI Will Now Pay Up to $172,000 for the Worst ChatGPT Bugs appeared first on DailyAI.

Google’s Dreambeans Reaches India After US Downloads Fell 84%

About 5,500. That is roughly how many times Google’s AI story app Dreambeans was downloaded in the US in October, according to Sensor Tower estimates reported by TechCrunch, down 84% from September. On Tuesday Google opened the app to everyone aged 18 and over in India, its first market outside the US.

Dreambeans is an experimental app from Google Labs, the company’s testing ground for new products. In its announcement, Google said it builds “personalized daily story collections” from the Google apps you choose to connect: Calendar, Gemini, Gmail, Photos, Search and YouTube. You don’t type a prompt. The app reads your accounts and writes illustrated stories about your plans, interests and memories, and some now come with short animations made by Google’s Veo video model.

In India it runs on Android and iOS in 14 languages, English plus 13 Indian languages including Hindi, Bengali, Tamil, Telugu and Urdu.

A slow start in the US

The numbers from home are modest. Sensor Tower estimates Dreambeans has been downloaded about 55,000 times in the US since it appeared in June, TechCrunch reported. About 63% of those came in September, when Google opened it to all eligible US adults.

In its first 30 days of full US availability, the app picked up an estimated 29,000 downloads. Meta’s Muse managed 7.4 million in its first 30 days, ChatGPT 2.8 million and Claude 155,000, on the same Sensor Tower estimates. Those are general chatbots, though, and Dreambeans does one narrower job.

Google product manager Gozde Oznur said India was picked because of strong demand and because many people there use Gemini, Gmail and Google Photos together, which gives the app more to draw on.

What it reads, and what you can switch off

The app’s whole idea is that it looks through personal data so you don’t have to ask. Oznur said users can choose which services to connect, disconnect them at any time and delete their Dreambeans data in the app. She said stories are visible only to the account holder, and the app works only with personal Gmail accounts, not work ones.

She also acknowledged that an app that digs up memories on its own could produce stories about people who have died. Users can tell it to avoid certain subjects, she said.

There is a practical limit for Indian users too. Stories come in one language at a time, so the app can’t write in the mix of Hindi and English that many people use every day. Google isn’t alone in pushing assistants deeper into personal data: Apple’s Siri now draws on your voicemails and past chats.

Why it matters

Dreambeans is a test of whether people want AI that acts on their data without being asked. So far few Americans have taken it up, and India, with a far larger base of Google users, is where Google will find out if that changes.

The post Google’s Dreambeans Reaches India After US Downloads Fell 84% appeared first on DailyAI.

OpenAI Says It Will Pay Its Own Power Bill. Will Yours Go Down?

Up to 1,000 megawatts. That is how much power OpenAI says its planned Georgia campus can give back to the grid when homes need it most, roughly the average demand of 700,000 households. On Monday the company made that deal the template for a national scheme it calls Grid Protectors, and promised that families won’t pay for its electricity.

The pitch lands at a sensitive moment. Americans are turning against data centres, in large part because of what they fear the buildings will do to their power bills. So the real question for most readers is simple. If an AI company moves in next door, does your bill go up or down? The Georgia numbers give a partial answer, and they come with caveats worth reading closely.

What OpenAI announced

In its announcement on 2 November, OpenAI said it will work with utilities, regulators and policymakers so that its future sites can “reduce electricity demand when the grid is under strain”. It also wants to supply power from on-site generators or batteries where local rules and equipment allow.

The idea has a name in the power industry: demand response. A big customer agrees to cut its use when the grid is stretched, for example on the hottest afternoon of a heatwave. In return, the utility doesn’t have to build as many new power plants just to cover those few peak hours.

OpenAI also made a promise about who pays. “Families and small businesses should not be asked to subsidize a compute campus’s electricity needs,” the company wrote. “Compute campus” is OpenAI’s new name for its data centres, part of a wider {I(SPEED,’effort to win back public opinion’)} ahead of a policy push in Washington.

The company gave no list of new sites, no new megawatt figures and no dates. It said arrangements “will differ by location” and that it will share more commitments “as they are finalized”. For now, the programme rests on one agreement.

The Georgia deal behind it

That agreement covers OpenAI’s planned campus in Effingham County, about 45 minutes from Savannah. When Georgia Power announced the project in July, it said the site is expected to need about 3,200 megawatts. OpenAI agreed to provide up to 1,000 megawatts of flexible demand under a 25-year contract, during which the utility can cut the power it delivers to the campus at certain times.

A plain reading: at most, a little under a third of the site’s expected load can be switched off when the grid needs it. That is a ceiling written into a contract, not something anyone has seen work yet. OpenAI itself says the campus is “still in development”.

Under the contract, OpenAI pays the full cost of the power lines, substations and other equipment needed to serve the site. That follows rules Georgia’s Public Service Commission (PSC), the state body that sets power prices, adopted in 2025 for customers that use at least 100 megawatts. Commission staff declined to object to the contract in August, which let it go ahead without a vote by the five commissioners. The contract itself has not been made public.

The $15 a month question

The number most households will care about comes from Georgia Power, not OpenAI. In its August release, the utility said its portfolio of big-customer contracts, OpenAI’s included, should bring in extra revenue worth about $950 million a year from 2029. It said that works out to savings of at least $15 a month, or $180 a year, for a typical home using 1,000 kilowatt-hours a month. That was up from a promise of $102 a year made in December 2025.

The logic is the one OpenAI leans on. Power lines and substations cost the same whether they run full or half empty. A customer that buys a lot of electricity all year, and steps back at peak times, spreads those fixed costs over more sales. Everyone else’s share of the bill gets smaller.

There are three caveats. First, the $950 million covers all of Georgia Power’s large new customers and “projected future growth”, so it isn’t OpenAI’s contribution alone. Second, as WJCL reported, the $15 “is not a bill credit or a guarantee that bills will decrease”. It means bills should sit at least $15 lower than they otherwise would have from 2029 to 2031, while fuel and storm costs can still push them up. Third, it depends on enough of those contracts being approved and the sites actually being built.

Georgia did add some protection for the case where a project fails. If a big customer walks away before its contract ends, Georgia Power must tell the commission within 15 days and has agreed not to ask homes to cover the shortfall through future rate rises, according to the staff letter WJCL described.

Why the timing matters

Public opinion has moved fast. In a Pew Research Center survey of 10,548 US adults carried out from 20 July to 9 August, half said data centres are mostly bad for home energy costs, up from 38% in January. Only 4% said they have a mostly good effect on those costs.

Politicians have noticed. Several states, including Tennessee, Florida, Oklahoma and Oregon, have passed laws meant to stop AI companies passing their energy costs on to residents, The Verge reported. Outside the US, Australia is planning to keep AI data centres out of its suburbs.

Electricity prices are also on the ballot in Georgia itself. The PSC is elected, and WJCL reported that Commissioner Peter Hubbard, who said he and fellow commissioner Alicia Johnson pushed Georgia Power into the bigger $15 promise, faces a rematch in this month’s election against the Republican he beat in 2025.

Then there is the scale of what is coming. OpenAI cites a forecast that US data centre capacity will pass 100 gigawatts by 2030. The most recent BloombergNEF outlook, from July, puts it at 118 gigawatts, a 52% rise on its forecast from last December. For comparison, the whole Effingham site at 3,200 megawatts would be 3.2 gigawatts of that.

What it doesn’t settle

Flexibility only helps if the campus really does power down when asked. A demand-response promise is easy to sign and can be harder to honour when there is money to be made from running chips flat out. Georgia Power’s twice-yearly reports on its data centre contracts, which it agreed to publish, will be the first public check on whether the sites are operating as promised.

OpenAI is also candid about the limits. “Flexibility cannot replace investment in new energy supply and transmission,” it wrote. Other customers may still end up paying for new power plants and lines that serve the AI boom, even if they don’t pay for the campus’s own connection.

And Georgia is one state with one set of rules. OpenAI’s Stargate sites, built with Oracle and SoftBank in Texas and New Mexico, sit under different regulators. Grid Protectors has no published terms for any of them yet.

Why it matters

If you live near a planned AI campus, the deal on the table matters more than the company’s slogan. Georgia shows what a written promise can look like: the tech company pays for its own connection, agrees to power down at peak times, and the utility puts a projected saving in writing. It is still a projection, and nobody has yet tested a 1,000-megawatt switch-off in a real heatwave.

The post OpenAI Says It Will Pay Its Own Power Bill. Will Yours Go Down? appeared first on DailyAI.

UK Regulators Say Firms Can’t Blame Their AI Agents When Things Go Wrong

Four UK regulators have told companies that “the AI did it” will not work as an excuse. If an AI agent breaks consumer law while acting for you, the firm that deployed it is on the hook, and the competition watchdog alone can fine a company up to 10% of its global turnover.

The warning comes in a joint statement from the Digital Regulation Cooperation Forum, published on 30 October. The forum brings together the Competition and Markets Authority (CMA), the Financial Conduct Authority (FCA), the Information Commissioner’s Office (ICO) and the media regulator Ofcom.

Its subject is personal AI agents. These are assistants that do tasks for you rather than just answer questions: booking things, searching, sending emails, managing expenses or making purchases. Tools like OpenAI’s Dot agent in ChatGPT already work this way.

What the regulators said

The four bodies said they want “to be explicit that existing legal and regulatory frameworks apply to personal AI agents”. That covers competition, consumer and data protection law, online safety, financial services, payments and cyber security rules.

The key line is about blame. “An AI agent’s autonomy does not diminish the responsibility of organisations that deploy it to comply with applicable law and relevant regulation,” the statement says. “Where we identify conduct that we reasonably suspect breaches the law and risks harm to consumers or businesses, we will not hesitate to use our powers.”

The regulators named the risks they have in mind: fraud and scams, personal data being shared in ways users don’t expect, and misleading recommendations or decisions. They said they are working with the government to find gaps in the rules, as law firm Burges Salmon summarised.

The number behind the warning

The threat has teeth because of a law that took effect in April 2025. Under it, the CMA can decide for itself that consumer law was broken, without going to court, and fine a company up to 10% of its global turnover. The statement sets out no new rules or fines of its own.

The CMA had already published guidance on agentic AI and consumers in March, and updated that page to point to the new statement.

The timing matters. Big US stores and banks are already blocking outside AI shopping agents, citing security. The UK message is aimed at the other side of that fight: companies that let agents loose on customers’ behalf.

Why it matters

If an AI agent buys the wrong thing, leaks your data or steers you to a worse deal, UK regulators say the company behind it is responsible. That gives British users a clearer route to complain, before any agent-specific law exists.

The post UK Regulators Say Firms Can’t Blame Their AI Agents When Things Go Wrong appeared first on DailyAI.