Mumsnet Denies AI-Written Posts as Fake AI Ads Steal Ad Accounts and Fadell Faults Early AI Gadgets

The prompt appeared in response to a user named AmIAComplete, who had asked for 'a sanity check' about an awkward social interaction with their brother's family involving a neurodivergent uncle and a sick child. The response was titled AIBU Response Schema v.1.0 and instructed an AI to write an 'interpersonal reasonableness' post, 'give a clear, forum-style ruling plus brief actionable advice,' and included definitions of common Mumsnet abbreviations including AIBU, ND and DD. It said the AI should not diagnose anyone, demean neurodivergent people or imply a sick child owes social performance, and should prioritize compassion, boundaries and de-escalation.

Users called the prompt 'really creepy,' 'crazy' and 'so disappointing.' One said, 'We might as well just go and do something else then.' The episode fed speculation about Mumsnet's use of AI at a time when the site, founded in 2000, reported a 19% drop in monthly unique users in 2025 to 6.1 million and a fall of 600,000 posts. Its latest annual report said it was 'embedding AI tools to improve productivity and enhance the user experience.' The Guardian also cited recent research in Nature showing AI assistance in social media risks creating 'semantic garbage' and is seen by users as less informative, 'robotic' and 'generic.'

Mumsnet first denied using AI to write threads and said the prompt had come from a tool that suggests thread titles using an OpenAI model. Users rejected that explanation; some copied the prompt into an AI and found it generated a full post. 'We are not idiots,' said user MrsBennetsPoorNervesAreBack. 'Please can you give us a transparent explanation?' Mumsnet then said the prompt did not come from them and suggested it had transferred from AmIAComplete's computer. Users were unconvinced and speculated it had leaked from Mumsnet's internal system.

Founder Justine Roberts joined the thread and confirmed the prompt had come from its system, which sends users' drafts to OpenAI to suggest a title, although this time it sent a prompt for a whole reply. She said it 'looks like an error on their side' and added: 'We don't use AI to write threads or replies on Mumsnet.' Users said they were being 'fobbed off.' One asked why OpenAI would do something so specific if it had never been asked to do that from Mumsnet's side, adding that it seemed to have appeared in the wrong place rather than sprung out of thin air. Another said, 'I just don't know anymore what's real and what isn't sometimes!' Roberts told The Guardian she understood why it looked like it had appeared in the wrong place, but said nothing in Mumsnet's systems asks for, stores or uses anything like it. She said Mumsnet had never tested a tool that drafted posts but used AI for coding, advertising operations, conversational analysis and flagging potentially problematic posts to moderators. 'Mumsnet works because the people on it are real,' she said. 'Members trust that the advice they get comes from other parents, not a machine... This episode was a useful reminder of two things. First, these systems are less predictable than the people selling them like to suggest. Second, users are understandably highly concerned that Mumsnet remains written by humans.'

In a separate cybersecurity report, TechRadar said cybercriminals are offering fake AI-powered products for advertisers and marketing managers to steal Google business accounts. The campaign abuses major AI names including ChatGPT, Google Gemini, Claude, Manus and Muse. Security researchers Island published a report detailing the operation after operators left source code for earlier versions on a misconfigured public GitHub repository.

The fake sites offer marketing-related products supposedly powered by AI. 'Every brand gets its own pitch,' the researchers said. ChatGPT promises a Monday Google Ads brief; Gemini promises manager-account and linked-client support; Claude gets an advertising portal; Perplexity offers campaign planning and spend audits; and Manus offers a private Meta integration. Each site asks the victim to connect their Google account, then launches a browser-in-the-browser attack that displays an entire browser, address bar and all, within the real browser. Victims might see a legitimate domain such as accounts.google.com in that fake address bar while the actual domain above has nothing to do with Google or OpenAI.

The phishing kit adapts to the visitor's operating system and includes details such as Safari's URL pill, Chrome's custom tabs and dark mode. A human operator follows the victim through the login process, sees each submission and decides what the victim sees next, including fake mistyped-password errors or a chosen MFA screen. The device is fingerprinted from IP and location down to screen size and WebGL. The platform supports Google, Meta, TikTok and Okta workflows, according to the report.

Island said the crooks deliberately target business accounts because they are connected to credit cards. 'The lures are written for agency staff, media buyers, and manager-account administrators, because an advertising account is a spending account,' the researchers said. 'It carries a stored payment method and an approved budget, and a manager account can reach several client accounts, each with its own billing profile and linked users.' Island could not disrupt the operation, which was still ongoing, and researchers saw 'hundreds of victim submissions to the platform.' They urged organizations to treat fictional AI integrations as account-access requests, inspect the outermost origin, correlate the client pattern and hunt for control vocabulary. They also recommended phishing-resistant authentication and reviews of advertising control changes. 'Origin-bound passkeys and hardware-backed authentication remove the reusable password and one-time-code material this platform is built to collect,' they said. After exposure, they said, organizations should check every client account the identity could reach for new managers or partners, changed recovery details and campaigns or spend nobody approved.

At the inaugural MIT Future Fest, Tony Fadell, credited as the 'father of the iPod,' a co-creator of the iPhone and founder of Nest, projected a slide with three discontinued AI devices: the Rabbit R1, the Humane AI pin and the Limitless pendant, TechCrunch AI reported. Fadell said these were the first-generation AI products, and companies called him for help because of his background, but he did not help.

Fadell said the devices failed to meet any kind of need. He said entrepreneurs have to understand what they are trying to do and what pain they are trying to solve. In each case, he said, the product did not meet a need; it was interesting technology for geeks that did not really apply to most people's lives. The early AI devices promised the convenience of a personal assistant but did not work well, and most people have never had an assistant, he said. He said less than 0.01% of the world population has ever had a human assistant do anything, and most consumers do not even know what an assistant is.

Fadell said building trust in an assistant takes years, including with sensitive data, banking and scheduling. He said trust and safety would be paramount for AI, and that Apple may be the only company positioned to deliver it because it has the hardware, chips and pieces, though not all the AI technology. He predicted a successful AI agent would have to operate on-device only for privacy and to keep the tech lightweight, and he rejected the idea that data centers will conquer the world. He noted Apple has earned trust to share biometric data through Face ID, but unlike other top tech companies it lacks a world-class AI model of its own; the new Siri AI runs on custom-built versions of Google's Gemini. Fadell theorized that companies such as Meta and OpenAI are turning to gadgets because they do not have billions of devices in circulation like Apple and lack access to phone sensors. They need video, audio, GPS and more, he said, so they make a screenless device with those sensors, connect it to phones and the web over Bluetooth, Wi-Fi or 5G to collect sensor data.

Editor's Summary

Mumsnet denied using AI to write forum threads after a prompt for a full reply appeared on its message board, and its founder later confirmed the prompt came from a title-suggestion system that sends drafts to OpenAI. Separately, researchers detailed a phishing campaign that impersonates AI brands to steal advertising accounts, and Tony Fadell said first-wave AI gadgets failed because they did not solve real consumer needs. The three developments underscore disputes over AI-generated content, security risks in AI-themed marketing, and skepticism about early AI hardware.