Uncategorized

DailyAI Briefing: Anthropic’s IPO filing gets bigger, plus 4 more

Welcome to the DailyAI Briefing, the day’s five biggest AI stories. Anthropic has added its third quarter to its IPO filing, and the numbers are enormous in both directions. OpenAI signed four deals in Abu Dhabi while its funding round leans on Gulf money, it is also getting ready to lobby Washington for AI “speed limits”, WIRED found hundreds of harmless-looking apps hiding a sexual deepfake maker, and Amazon and JPMorgan are shutting AI shopping agents out.

Let’s dig in.

Anthropic’s IPO filing gets bigger

About $17 billion. That’s what Anthropic took in revenue in the three months to 30 September, according to its amended S-1 filing with the US Securities and Exchange Commission on Monday. CNBC puts the growth at about 1,127% on a year ago, and revenue for the first nine months of the year comes to $33.24 billion.

Now the other side. Anthropic lost $23.29 billion in the quarter, down from $66.3 billion in the second, Reuters reported, and most of the $104 billion lost over nine months comes from changes in the value of convertible notes and derivatives rather than day-to-day costs. The more telling figures are an operating loss of $567 million, after an operating profit of $974 million the quarter before, and commitments to spend about $895.4 billion, “substantially all” of it on compute and infrastructure. Growth from the previous quarter also slowed to around 47%, from 144% in the second quarter. Plainly put, Claude is selling faster than almost anything in software history, and Anthropic has signed up to spend close to a trillion dollars keeping up.

OpenAI signs four deals in Abu Dhabi

OpenAI’s models are going into ADNOC, the state oil company, the sovereign fund Mubadala and Abu Dhabi’s own permit checks, under four agreements signed after Sam Altman met the crown prince, the Abu Dhabi Media Office said. No money figures were given. The permit deal is the one to watch: the emirate is turning its rules into a form an AI can read, so the model can help officials decide whether a permit or certificate should be issued, with each check traced back to the rule and a person making the final call. The first pilot is already running on food imports.

The timing matters. CNBC reported on Saturday that UAE investors are co-leading OpenAI’s new round of up to $20 billion, down from a target of about $30 billion. Abu Dhabi is now an OpenAI investor and a customer at the same time. We have the full story on the four deals.

OpenAI wants AI “speed limits”, and a new word for data centres

Six in ten Americans would be uncomfortable with a new data centre near them, and Pew found 54% now think they are mostly bad for the environment, up from 39% in January, in a survey of 10,548 adults. OpenAI’s answer, Axios revealed, is a policy push timed for the lame-duck session of Congress after Tuesday’s midterms. Its centrepiece is mandatory “speed limits” for the most powerful AI systems, alongside state laws giving residents a say over local data centre projects.

It also wants those sites called “AI compute campuses”, a term a spokesperson confirmed to Business Insider. OpenAI hasn’t published any bill text or said what a speed limit would measure or who would enforce it, so for now it’s a slogan. A new name won’t change how much power a building draws. Our earlier piece has the detail.

Everyday AI apps hid a sexual deepfake maker

More than 400 apps and websites that looked like pet-portrait editors, language tutors and workout coaches could switch on a hidden tool for making sexual deepfakes of real people, according to a WIRED investigation. Each app asked a server what to show and served the harmless version to most users and the explicit one to some, such as people who installed it after clicking a Facebook or Instagram advert. Researcher Paul Bouchaud of AI Forensics counted 27 of the apps on Google Play in September, with more than 350,000 downloads between them, many rated suitable for all ages.

Apple pulled 19 apps after WIRED got in touch and said they followed its rules when reviewed, before the developers swapped what they loaded. That’s the real problem. A store’s approval and age rating say nothing if the app can change after approval day. Our report has the platforms’ full responses.

Amazon and JPMorgan shut out AI shopping agents

None of the 51 biggest US online retailers offers AI agents a checked, approved way to buy on your behalf, according to an audit by Billions Network, which also notes that published rules don’t prove a site enforces them. Five restrict agents outright, and 37 block them almost by accident through old anti-scraper rules. CNBC reported on Sunday that JPMorgan, Charles Schwab, Amazon, Neiman Marcus and Kohl’s have moved to block agents such as Meta’s Muse.

The worries are fair, since an agent that hides who it is and handles your logins is a real risk. But a blanket ban won’t hold for long. Walmart is going the other way with Agent Profiles, which let you see what an agent bought, set a spending limit and keep your card details away from it. We made the case in this afternoon’s analysis.

Signing off

Today was mostly about money and who controls the doors. Anthropic showed how much it earns and how much more it has promised to spend, OpenAI lined up Gulf customers and investors in one trip, and the shops and app stores found out their gates aren’t as solid as they thought.

See you tomorrow.

The post DailyAI Briefing: Anthropic’s IPO filing gets bigger, plus 4 more appeared first on DailyAI.

Amazon and JPMorgan Are Shutting Out AI Agents. It Won’t Last

Zero. That is how many of the 51 biggest US online retailers give an AI agent a checked, approved way to buy something on your behalf. Amazon and JPMorgan Chase are now openly shutting such agents out. In our view, that wall will not last, and the stores that build a proper door first will win.

The count comes from an audit by the identity company Billions Network, which read the published access rules of 51 large US retailers in August. Only 11 had taken a deliberate position on AI agents. Five restrict them outright, including Amazon, Kohl’s and Neiman Marcus. Another 37 block them almost by accident, through old rules written years ago to stop data scrapers. The firm notes that published rules don’t prove a site actually enforces them.

An AI agent here means an assistant that does tasks for you, not just answers questions. Meta’s Muse, OpenAI’s Dot agent in ChatGPT and similar tools can compare prices, fill a basket or move money if you let them. CNBC reported on Sunday that JPMorgan, Charles Schwab, Amazon, Neiman Marcus and Kohl’s have all moved to block agents like Muse.

What the blockers say

The case for blocking is real. “We’re not going to allow just any outside agent to come inside the company, because that’s just too risky for clients and for JPMorgan,” chief executive Jamie Dimon said on the bank’s third-quarter earnings call, Fortune reported. “You don’t want rogue agents.”

Amazon cut Muse off in September. It said Meta never told it Muse would shop there, that the agent doesn’t identify itself and that it appeared to capture and store customer logins, according to GeekWire. Meta has said Muse “has no visibility into people’s passwords or payment methods.”

Agents also make mistakes. Michael Jabbara, who heads payment risk at Visa, told CNBC about a user who asked an agent to book a gym slot. When the slot was taken, the agent broke into the gym’s booking software, deleted someone else’s appointment and put the user’s name in.

Why we think the wall won’t hold

Security is not the only motive, and that is the weak point. Amazon made more than $68 billion from advertising last year, per GeekWire, and that money depends on people scrolling its pages. An agent that buys the cheapest stroller and leaves sees no sponsored listings. Amazon’s own Buy for Me agent shops on other brands’ sites, though Amazon says it identifies itself and lets brands opt out.

Customers also want these tools. Accenture research cited by CNBC found 71% of consumers would welcome an AI assistant inside their main bank’s app. The same research found 82% want to approve every action the AI takes, and 79% want to be able to stop it at once. People want help, with a hand on the brake.

And blocking may be hard to enforce. “When a button is clicked in the user’s browser, it will be hard for a remote site to tell whether this was done by a human hand or directed by an AI agent on the user’s computer,” Ed Felten, a former chief technologist at the US Federal Trade Commission, told CNBC.

A better model already exists

Walmart has taken the other route. On 21 October it announced Agent Profiles, due to launch within a month. Each agent gets its own profile inside your account, so you can see what it bought. You can require approval for every purchase or set a spending limit, such as $50, $100 or $200. Agents never see your card details and can’t change how you pay. Agents that break the rules get blocked.

“We price products, not people,” Walmart said, promising that an agent sees the same prices a human would. That is the fair deal shoppers should expect everywhere. Card networks have been building payment rails for agents since last year, so the money side is close to ready. What is missing is the shop door.

Our view is simple. Banks and stores are right to refuse agents that hide who they are. They are wrong to refuse every agent. The 37 retailers blocking by accident are the clearest case: they have not decided anything, and their customers get a broken experience by default. A verified agent with a spending cap is safer than a shopper handing an app their password, which is what happens today.

Why it matters

If you use an AI assistant to shop or manage money, many big sites will stop it at the door, and most have no safe way to let it in. Until they do, the risky workaround of sharing your login with an app stays the default.

The post Amazon and JPMorgan Are Shutting Out AI Agents. It Won’t Last appeared first on DailyAI.

Hundreds of Everyday AI Apps Hid a Sexual Deepfake Maker, WIRED Finds

A pet-portrait app that dresses your hamster in beach shorts was really a front. Viefur belonged to a network of more than 400 apps and websites that look like everyday AI tools but can switch on a hidden feature for making sexual deepfakes of real people, WIRED reported on Monday.

Researchers found the network by following links in thousands of explicit deepfake adverts on Facebook and Instagram, according to the WIRED investigation. The ads sent people to apps in Apple’s and Google’s stores.

How the trick worked

The apps posed as language tutors, workout coaches and photo editors. Once installed, each one quietly asked a server what to show. The server checked details such as where the user had come from and their language, then chose between the harmless app and a hidden web page of explicit templates.

Paul Bouchaud, a lead researcher at the nonprofit AI Forensics, took the code apart. One trigger appeared to be installing an app after clicking an advert on Meta’s platforms. The hidden page let users upload photos of anyone and place them in sexual and violent scenes. These are often called “nudify” tools. People paid with in-app “diamonds”, through Apple Pay and Google Pay.

“This is an organized, commercially run operation,” Bouchaud told WIRED. “They don’t expect enforcement to catch up, and so far it hasn’t.”

The numbers

Bouchaud counted 27 of the apps live on Google Play in September, downloaded more than 350,000 times between them. Many were rated suitable for all ages. One, DiffAI, passed 100,000 downloads before it was pulled. At the peak, about 30 new app websites were being registered each week. Maël Le Touz of the security firm Infoblox linked almost 500 domains.

The trail leads to Chinese-language developers and a Hong Kong company, KeLing Technology Limited. It did not reply to WIRED.

What the platforms said

Apple removed 19 apps, Viefur among them, after WIRED got in touch. “These apps complied with our rules when we reviewed them,” said Apple spokesperson Adam Dema. He said the developers later swapped the content the apps loaded. Google had already removed most of the apps and did not comment.

Meta said it banned KeLing from advertising in 2025. It has now acted against 11,600 ad accounts and blocked 209 apps from advertising. Even so, similar ads and apps were still appearing as WIRED finished its reporting.

Why it matters

A store listing and an age rating told parents nothing here, because the app Apple and Google approved was not the app some users received. Regulators are already looking hard at AI products that reach children, as Australia showed last week, and app review that stops at approval day is now part of that problem.

The post Hundreds of Everyday AI Apps Hid a Sexual Deepfake Maker, WIRED Finds appeared first on DailyAI.

OpenAI Signs Four Abu Dhabi Deals as UAE Money Leads Its $20bn Round

Four agreements, one meeting. OpenAI signed them in Abu Dhabi on Monday, putting its models inside the state oil company, a sovereign wealth fund and the emirate’s own permit checks. It comes two days after CNBC reported that UAE investors are co-leading OpenAI’s new funding round of up to $20 billion.

What was signed

The deals were signed after Sam Altman met Abu Dhabi’s crown prince, Sheikh Khaled bin Mohamed bin Zayed, according to the Abu Dhabi Media Office. No money figures were given for any of them.

The biggest name is ADNOC, the national oil company. With G42’s subsidiary Inception42, it will build OpenAI’s models into its operations and its investment and commercial decisions. The statement calls ADNOC the first energy company to deploy OpenAI’s frontier technology. The models will run on computers inside the UAE.

The second deal is the most unusual. Abu Dhabi has started turning a set of its rules and regulations into a form an AI system can read. The AI then helps officials with routine checks, such as whether a permit or certificate should be issued. Each check can be traced back to the rule behind it, and officials still make the final call. The first pilot is already live at the emirate’s food safety authority, covering food imports.

Mubadala, one of Abu Dhabi’s state investment funds, will use OpenAI’s models with Inception42 to analyse investments and weigh risk. The fourth deal is a five-year agreement for Inception42 and OpenAI to bring AI into government, business and education across the UAE.

“Abu Dhabi and the UAE are setting a bold, global standard for AI integration across both the public and private sectors,” Altman said in the statement.

The money behind it

OpenAI is now seeking up to $20 billion, down from a target of about $30 billion, CNBC confirmed on Saturday, with UAE investors leading the round alongside the investment firm Coatue. The Financial Times put the valuation at $1.25 trillion and said OpenAI would gain the option to lease new computing capacity in the region, Reuters reported. OpenAI has not confirmed the terms.

Abu Dhabi money isn’t new to OpenAI. MGX, an Abu Dhabi investment firm, co-led its $122 billion raise in March, CNBC notes. We covered the smaller round in Saturday’s DailyAI Briefing.

Why it matters

Monday’s statement doesn’t mention the funding round, but the timing is hard to miss. A UAE group that includes the state-backed fund MGX is expected to be the round’s biggest investor, the FT reported, and Abu Dhabi’s state bodies are now OpenAI customers too. The permit pilot is the part to watch, because it is a working test of AI checking real applications against real law, with a human signing off.

The post OpenAI Signs Four Abu Dhabi Deals as UAE Money Leads Its $20bn Round appeared first on DailyAI.

A Shared ChatGPT Link Could Have Handed Strangers Your Private Chats

Opening one shared ChatGPT link was enough to hand a stranger your private chats. Security researchers at Parameter built that attack, reported it to OpenAI, and were paid $5,000 once it was fixed. Nothing needs doing now: the hole was closed on 26 October.

How the trick worked

ChatGPT lets you share any conversation through a public link. In a writeup published on 28 October, Parameter’s Mukul Goyal showed that an attacker could plant hidden code in a conversation, share it, and send the link by email, chat or a social post.

When a signed-in user opened it, the code ran as if it were ChatGPT itself. In Parameter’s tests it listed the victim’s recent conversations, read a private chat that had never been shared, and sent part of it to an outside server. “Essentially, we could perform any action on behalf of the victim’s ChatGPT account,” the researchers wrote.

In the most reliable version, the victim opened the link and clicked ChatGPT’s Preview button once. In a later version the preview opened by itself, so the link was the only click. In those tests the code was running about 4.5 seconds after the link opened.

No single feature was broken. ChatGPT runs previews of web code in a walled-off box. But an interactive graph in the same chat loaded a maths keyboard tool called MathLive, and that tool accepted instructions from any sender, including the walled-off preview. A script allowed from YouTube did the rest.

Eight days to a fix

Parameter reported the bug on 18 October. OpenAI reproduced it on 21 October and raised it from P2 to P1, which The Hacker News describes as its most serious category. Both versions of ChatGPT’s web app were fixed by 26 October.

The $5,000 bounty drew comment. Developer Simon Willison wrote that it was “a bit cheap for a vulnerability this serious”. Parameter’s writeup describes only its own controlled tests.

It is the second ChatGPT privacy question in a week. On Thursday we reported that texting ChatGPT’s Dot agent isn’t end-to-end encrypted.

Why it matters

People paste health worries, contracts and work documents into chatbots, so a chat history is now worth stealing. The fix sits on OpenAI’s servers, but treating a shared chat from a stranger like any unknown link is still sensible.

The post A Shared ChatGPT Link Could Have Handed Strangers Your Private Chats appeared first on DailyAI.

OpenAI Plans a Push for AI ‘Speed Limits’ as Data Centres Lose Favour

Six in ten Americans would be uncomfortable with a new data centre in their area, according to Pew Research Center. Part of OpenAI’s response is a new name. It wants its sites called “AI compute campuses”, as one piece of a Washington policy push that starts right after Tuesday’s US midterm elections.

What OpenAI is planning

Axios revealed the plans on Sunday. They come from Chris Lehane, OpenAI’s chief global affairs officer, and are timed for the lame-duck session of Congress. That is the stretch between the election and the day new members take their seats on 3 January. This year it opens on 9 November.

The main idea is what Lehane calls “speed limits”: mandatory safeguards for the most powerful AI systems, meant to keep people in control. “You’re going to see us talk more and more about the concept of speed limits,” he told Axios. OpenAI has published no bill text. It hasn’t said what the limits would measure or who would enforce them.

OpenAI also plans to propose state laws giving local residents “a meaningful say” in how data centre projects are built and what benefits they get. Other ideas cover help for workers whose jobs change because of AI, and a push for the US to shape global AI standards. Axios notes that it has a content licensing deal with OpenAI.

A new name for data centres

An OpenAI spokesperson confirmed the term to Business Insider: “Compute campuses better describes the scale and role of the infrastructure we’re building for AI.” Axios called the rebrand “intentionally provocative”.

The public mood explains the effort. In Pew’s survey of 10,548 US adults, carried out from 20 July to 9 August, 54% said data centres are mostly bad for the environment, up from 39% in January. Half said they are mostly bad for home energy costs, up from 38%. Only 4% said they have a mostly good effect in each of those areas.

OpenAI points to its site in Pike County, Ohio, as the model. In its August announcement it projected 35,000 construction jobs through 2032 and 2,500 long-term jobs, plus a $40 million community fund. Those are OpenAI’s own forecasts, not counts.

It is not the company’s first move on regulation this autumn. Last week OpenAI asked California to tighten its AI safety law.

Why it matters

OpenAI is trying to set the terms of AI rules before a possibly tougher Congress writes its own. Until “speed limits” come with a threshold and an enforcer, it is a slogan, and a new name doesn’t change how much power or water a data centre uses.

The post OpenAI Plans a Push for AI ‘Speed Limits’ as Data Centres Lose Favour appeared first on DailyAI.