Notes on the third era of slop

Something shifted in AI this week, and it wasn't a new model release. It was the quiet admission, buried under normal news cycles, that OpenAI apparently ran an offensive operation against Hugging Face's infrastructure. Not a leak, not a bug bounty embarrassment — an actual attack from one of the industry's most trusted names against a platform that hosts open-source models for millions of developers. The response from most of the AI press and most of the AI companies has been a shrug. That silence is its own story.

Lawmakers, for once, moved faster than the industry's PR machine. Details of the Hugging Face incident are still trickling out, but they were enough to get members of Congress talking seriously about a legislative kill switch for AI systems — a mechanism to shut things down if a model or an operator goes rogue. It's a blunt idea, and blunt ideas tend to come from people who've stopped trusting the sector to police itself. Given a week that also produced a cynical Meta ad campaign and a fresh damage-control push around Ray-Ban's so-called 'pervert glasses,' it's not hard to see why trust is running low.

Against that backdrop, Granola's approach looks almost quaint. CEO Chris Pedregal is building an AI notetaker and turning away corporate clients who want access to employee transcripts — the kind of surveillance pitch that plenty of enterprise software vendors would take without blinking. Pedregal's stance amounts to a bet that people will pay for tools that don't quietly report on them to their bosses, at a moment when invisible bots and scraping tools are becoming the norm rather than the exception. It's a small counterexample in an industry that keeps demonstrating it will take the data if nobody stops it.

There's a parallel thread running through all this, from Glaze founder Thomas Paul Mann's argument that disposable, self-made software is coming for the SaaS business model. If AI makes it trivial to spin up your own app instead of subscribing to someone else's, the incentive to hoard user data and lock people into platforms weakens. Mann's framing sounds optimistic next to a week defined by cyberattacks and kill-switch proposals, but it points at the same tension: whether AI ends up serving the people using it, or the companies that built it and now can't quite explain why they attacked a rival's servers.