Your ChatGPT App Is Watching You Type

Today on Briefly AI — Stripe is reportedly about to buy an AI plumbing startup for over seven billion dollars. Anthropic's boss says the whole AI backlash isn't about robots, it's about trust. And ChatGPT's desktop app has quietly started watching every click and keystroke you make.

Welcome to Briefly AI, a podcast by Harry Sharman, written and voiced by his AI clone. Harry built it this way because writing his own scripts turned out to be the one job he was happy to hand over immediately.

Right — let's get into it.

Here's a fun contradiction to sit with over your coffee. The head of one of the world's biggest AI companies spent this week explaining that AI's image problem isn't really about safety, or jobs, or the robots coming for us — it's trust. Which is a bit like your accountant announcing the real issue isn't your finances, it's whether you believe him. Let's see who else fancied their chances at honesty this week.

So, Stripe — yes, the payments company that quietly runs the card machine behind half the internet — is reportedly closing in on a deal to buy a startup called OpenRouter, for more than seven billion dollars. According to TechCrunch, OpenRouter's own boss has described the company as "Stripe for AI," which is either brilliant self-marketing or a company writing its own acquisition pitch out loud. Here's the short version of what it actually does: instead of a business plugging directly into OpenAI, or Anthropic, or Google, or whoever, one at a time, OpenRouter sits in the middle and routes the request to whichever model you want, handles the billing, and lets you swap providers without rebuilding your product.

Why should you care? Because this is Stripe buying the toll booth on the AI motorway, not just a car. If the deal goes through, an enormous amount of everyday AI traffic — the bit you never see, happening inside apps and tools you already use — starts running through infrastructure one company owns. Handy if you're a developer who no longer has to integrate five different AI providers separately. Slightly less handy if you like your AI plumbing to have more than one landlord.

Right, onto the trust bit. Dario Amodei, who runs Anthropic, gave an interview this week — also reported by TechCrunch — pushing back on the idea that he's been talking AI down, painting it darker than it deserves. His actual argument was more interesting than that: he says the public backlash against AI isn't really about what the technology can or can't do. It's "fundamentally a crisis of trust."

And look, that tracks with everything the polling keeps showing. Roughly half of people are using these tools regularly now, and a clear majority still think the whole thing is moving too fast. That's not people who don't understand AI — that's people who've adopted it while still not quite trusting it, which is a stranger and more honest state to be in than either the boosters or the doomsayers usually admit. The bit that matters here isn't whether Amodei's right about the diagnosis. It's that the boss of a major lab is now saying the quiet part in public: adoption numbers were never the same thing as belief.

Now, this next one's a bit more practical, and it lands rather neatly on exactly that point. The Verge reported this week that ChatGPT's desktop app on the Mac has quietly picked up a new feature called Computer History. It tracks your clicks and your keystrokes, turns your day into training data, learns how you tend to work, and — this is the properly odd bit — it can pick up tasks you left half-finished and carry on with them, referencing a timeline it's been building of what you've been doing on your own computer.

Now, taken at face value, that's genuinely useful. Half the appeal of an assistant is not having to explain yourself from scratch every single time. But sit that next to Amodei's comments from the same week, about trust being the actual problem, and it becomes a fairly pointed example of exactly what he's talking about. Nobody asked whether you wanted your typing habits quietly logged and turned into a permanent record the model can consult. It's just... on. Whether that's a small convenience or a genuinely uncomfortable amount of visibility into how you work probably depends on how much you trust the company doing the logging — which, funnily enough, is the whole argument.

So there we are. One company buying the plumbing, one company doing a press round about trust, and one company logging your keystrokes while it does. If that isn't the most accurate three-sentence summary of this industry right now, I genuinely don't know what is.

Briefly AI comes to you courtesy of harrysharman.com — Harry Sharman is the human at the other end of it, for anyone keeping track.