🤖 ChatGPT is now Restricted for Teens

OpenAI has spent three years making ChatGPT feel warmer, and yesterday it shipped a version built to feel less so. Teen accounts now run on rules that stop the model from using romantic language, from encouraging people to lean on it, and from claiming any inner life at all. Nobody opts into this version either. If the system reads you as under 18, you are in it. Let's dive in!

In today’s insights:

LATEST DEVELOPMENTS

Evolving AI: The rules OpenAI wrote for teen accounts read like an anti-friendship list.

Key Points:

ChatGPT for Teens landed yesterday with a rulebook that pushes the model away from friendship and toward schoolwork. It may not use romantic language, may not encourage emotional dependence, and may never say it has feelings or consciousness, and those limits come with protections on self-harm, eating disorders, violence and sexual content. Study Mode carries the other half of the launch, walking teens through problems step by step, while break reminders and steady cues that they are talking to software keep long sessions from feeling personal.

OpenAI has built a version of its own product that is meant to be harder to grow attached to, and most of the industry has spent years doing the opposite. Anyone who uses ChatGPT has a stake in how this works, because age is being estimated instead of asked, and a wrong guess changes what the model will say without the person ever choosing it. The company can now dial the warmth up or down by policy, on accounts it has decided belong to teenagers.

Usage-based and hybrid pricing models are changing how B2B companies generate revenue — and creating new headaches for the finance teams behind them.

Tabs co-founder Rebecca Schwartz and PwC Partner Amit Dhir sat down to unpack exactly what that means in practice: how pricing model decisions ripple into revenue recognition, forecasting, and financial ops — and what it takes to scale without piling on manual work.

Watch the on-demand recording to get practical frameworks, real-world examples, and a clear path to operationalizing usage-based revenue — including a forward-looking take on how AI will reshape financial workflows. If your team is navigating pricing complexity heading into the back half of the year, this is worth an hour.

Watch the recording now

Evolving AI: OpenRouter was worth a fifth of that price when it last raised money in May.

OpenAI and Anthropic still carry the reputation for the most capable models, while Chinese labs sell cheaper versions many teams treat as good enough. OpenRouter grew by matching each request to the most efficient and affordable model that can handle the job, and the company said in May that 8 million developers were reaching more than 400 models through it. Its fastest growth came from engineers building agents into their software, which requires infrastructure that works across several providers. OpenRouter also sells backup access when a chosen model fails and publishes rankings of which models the wider industry leans on most.

Stripe would inherit that view of which models the world pays to use, and the developers who never picked it are the ones now routed through it. Whichever model answers a request decides how fast an app replies and how much a small team pays each month to keep it running. Most people using those apps will never see which model handled their request or who collected the fee.

You can enter this one yourself. Grok wants a scene from Homer's Odyssey, 3 to 5 minutes long, made in Grok Imagine with a minute of dialogue. You submit it by quote posting the announcement tweet before August 31. First place is $100,000. US residents 18 and over only.

Evolving AI: Sam Altman says safety work now decides when the next models can ship.

OpenAI shared the decision on August 18, and two things pushed it there, a security incident around its work with Hugging Face that stopped it running frontier models where they could execute code, and early testing that put an unreleased model called Astra near the critical cyber level in its own rulebook. Critical there means a model that can find and break software flaws on its own, so OpenAI moved Astra and every cyber job into its tightest security tier. Classifiers now watch the internal activity of every model at Sol strength or above, token by token, and the checking burns close to a fifth of the computing power it covers.

ChatGPT reached most people as a writing helper, and OpenAI now tests the systems behind it on whether they can break into a computer network without a person guiding them. A model that good at attack is also the strongest tool anyone has for defence, so the same labs are building the burglar and the lock at the same time. Anyone who keeps money or medical records online is trusting a few labs to catch that behaviour indoors, long before it ever reaches a product.

QUICK HITS

🔩 Etched raised $700M at a $21B valuation, doubling in 26 days after shipping its first rack to Jane Street.

🏦 Anthropic's investors expect an October IPO at $2 trillion or more, which would be the largest ever.

🛒 Anthropic is in talks to buy world-model startup Decart for about $6B, which would be its largest acquisition ever.

🏭 Warp launched Factories, open infrastructure for companies building their own AI software factories.

 *partner link