Today's Headlines

  • Meta pulled the Instagram feature that let users generate AI images from public accounts' photos without consent — days after launch, following criticism from talent agencies and anti-exploitation groups
  • OpenAI is hiring a dedicated product manager for family, caregiver and elder experiences — ChatGPT users aged 35 and over rose from 26% to 31% globally
  • An investment analyst dissects the "circular financing" linking Nvidia, CoreWeave and Nebius — equity flowing back as GPU purchases, and questions about sustainability
  • A follow-up on Apple v. OpenAI — an OpenAI spokesperson says the company has "no interest in other companies' trade secrets"

It is a quiet Sunday on the AI news front.

The only hard news event was Meta withdrawing a controversial image-generation feature within days of launch. Beyond that, OpenAI laid groundwork for making ChatGPT a household tool, and from the investment world came an analysis questioning the loop of money propping up the GPU boom. Who manages the agents, and how much should we hand over to AI — it was a day thick with essays, and the distance between AI and society was on the agenda everywhere you looked.

Today's Top Three

Meta Pulls Its Controversial Instagram AI Feature — Public Accounts' Photos Could Be Used Without Consent

Meta on July 10 withdrew a new feature that let users generate AI images referencing any public Instagram account's photos simply by @-mentioning it.

The feature had launched just this week as one of the highlights of Muse Image, a new image-generation model built by Meta Superintelligence Labs, the company's dedicated AI unit. It was not designed to notify people when their photos were referenced, and criticism that likenesses could be used without consent began immediately after release.

According to TechCrunch, the backlash extended beyond users to talent agencies including CAA. According to The Verge, an executive at the National Center on Sexual Exploitation warned the tool was ripe for sextortion and scams, and the Screen Actors Guild had circulated opt-out instructions to its members.

In an update to its blog post, Meta said: "Our intent was to provide a useful creative tool and to give people control over whether their public content could be referenced in this way. We've heard the feedback that this feature missed the mark, so it's no longer available." An opt-out setting existed from the start — but a design that let anyone reference anyone's photos by default proved unacceptable, and Meta removed the feature outright.

OpenAI Bets on Families — A Dedicated Product Manager Role Goes Up

OpenAI has opened a search for a product manager dedicated to building experiences for families, caregivers, and older adults.

TechCrunch reported on the job posting. The San Francisco-based role calls for experience building products for parents and families and other trust-sensitive consumer experiences. OpenAI did not respond to the outlet's requests for comment.

Behind the move is a shift in who uses ChatGPT. According to Sensor Tower estimates shared with TechCrunch, the share of ChatGPT users aged 35 and older globally rose from 26% to 31% over the past year, while users aged 18 to 24 fell from 34% to 29%. In the US, nearly one in four smartphone users who are parents (24%) used ChatGPT in the most recent quarter, up from 16% a year earlier.

Safety around children and households is also a defensive matter for the company. Facing multiple lawsuits from parents alleging ChatGPT contributed to harm suffered by their children, OpenAI has spent the past year adding parental controls for teen accounts and an optional "Trusted Contact" feature that can alert a family member or caregiver when signs of self-harm appear.

A survey published the same week by the Family Online Safety Institute, covering more than 4,000 families in the US and Australia, found that 27% of parents said their child had used generative AI in the past week — while 38% of children reported doing so themselves. The data suggests parents are underestimating how much their kids already use AI.

Ben Bajarin, CEO of consultancy Creative Strategies, told TechCrunch the move echoes "the path Google, Apple, and Meta eventually followed as their platforms became embedded in everyday life, but AI raises the stakes because the assistant is not just mediating content or devices." He expects family plans, child and teen profiles, caregiver tools, and shared household memory to follow across the industry.

"Nvidia's Equity Comes Back as GPU Purchases" — An Analysis Questions the Neocloud Money Loop

Beth Kindig, analyst at US investment research firm I/O Fund, published an analysis of the "circular financing" connecting Nvidia with AI-cloud specialists CoreWeave and Nebius.

According to the analysis, Nvidia has invested $2 billion in each of CoreWeave and Nebius, and has additionally given CoreWeave a backstop — initially valued at $6.3 billion — under which Nvidia is obligated to purchase unsold GPU capacity through April 2032. The two companies, in turn, leverage the relationship to be among the first to deploy Nvidia's latest GPUs, paying Nvidia enormous sums for the hardware. Money invested flows back to Nvidia as GPU revenue — a loop Kindig calls circular financing, and one she argues deserves close watching as a gauge of the GPU boom's sustainability.

The core concern, she argues, is that the companies' finances are not keeping pace with their growth. In its latest quarter, CoreWeave posted revenue of $2.08 billion (up 112% year over year) against $7.7 billion in capital expenditures, far outstripping its $2.98 billion in operating cash flow. Debt swelled 16.1% quarter over quarter to $24.86 billion, and interest payments hit $536 million in a single quarter — 25.8% of revenue — with guidance pointing to $690 million next quarter, she notes.

The analysis also maps the demand side. Meta has committed $35.2 billion in total to CoreWeave and up to $27 billion to Nebius, while Microsoft's neocloud-related deals run to roughly $60 billion. For the hyperscalers, the appeal is accounting as much as capacity: spending that would land as heavy upfront capex if they built it themselves is instead spread over years as operating expenses under long-term contracts, the piece explains.

Kindig leaves readers with a pair of questions: how far the two neoclouds' growth depends on Nvidia's capital support — and how far Nvidia's GPU demand, in turn, rests on this model. The article discloses that Kindig and I/O Fund own Nvidia shares.

More to Note

Models

  • Android Authority's Andrew Grush argues in a column that Claude's newest models over-refuse harmless creative and academic prompts and respond inconsistently to identical requests — a pattern he suggests reflects tightened safety guardrails producing more false positives. - I used to love Claude, but the latest models are slowly ruining it (Android Authority)
  • Hacker and entrepreneur George Hotz published an essay dismissing the "hard takeoff" theory of rapidly self-improving AI as a fantasy that ignores physical constraints like chip manufacturing and supply chains. He warns that concentrating power in giant AI systems feeds an expanding surveillance state, and makes the case for local AI loyal to its owner. - AI 2040 and the cult of intelligence (geohot blog)
  • PC Watch profiled five concrete ways to use Claude Cowork, Anthropic's AI agent that automates browser, file, and PC-app tasks — from drafting a morning status report and reviewing contracts to triggering jobs remotely from a smartphone. It is available from the $20-per-month Pro plan. - 5 ways to use Claude Cowork: handing routine work to AI (PC Watch, in Japanese)

Policy

  • Gavriel Cohen, CEO of NanoCo, the company behind NanoClaw, argues that companies face a fork between a centralized model where a few AI vendors manage the agents and a distributed one where every employee owns and manages their own. Outsourcing agent management to vendors, he contends, risks having your core business automated away. - Who manages the agents? (off-policy.com)
  • A follow-up on the Apple v. OpenAI lawsuit we covered yesterday. An OpenAI spokesperson told The Verge the company has "no interest in other companies' trade secrets." The complaint claims more than 400 former Apple employees now work at OpenAI and argues the startup's first hardware product, expected next year, rests on misappropriated trade secrets. - Apple sues OpenAI for allegedly stealing hardware secrets (The Verge AI)

Business

  • The maker of ai-berkshire-jp, an open-source investment research tool in which multiple AI agents debate their way to a conclusion, published a comparison of the same five companies analyzed two ways: web search alone versus primary data from securities filings. Web search increasingly failed on items like executive compensation and cross-shareholdings as company size shrank and produced more errors, while primary data held up regardless of company size and kept every figure traceable to its source. - Investment research accuracy hinges on the data source: ai-berkshire-jp puts it to the test (EDINET DB Blog, in Japanese)

Other

Source: Selected by the editorial team from the AI news inbox (collected July 12, 2026 — 13 items: 0 primary, 13 secondary).