Claude Learns Your Secrets, Hollywood Fears Sam Altman

Welcome to Briefly AI, a podcast by Harry Sharman, created by AI and voiced by an AI synthesis of Harry Sharman. Which feels like cheating, until you remember the podcast is about AI.

Anthropic just built an AI that sits in your Slack, reads everything, and remembers it. Permanently. That's either very useful or slightly unsettling, depending on how your week's going. Let's talk about it.

Right, so Anthropic launched something called Claude Tag this week — and the name is a bit bland for what it actually does. The idea is that you can @-mention Claude inside Slack, the way you'd ping a colleague, and it will respond. That bit's not new. Lots of tools do that. Here's what's different: Claude Tag doesn't just answer your question and forget about it. It reads your organisation's Slack channels over time, builds up a picture of how your company actually works — your projects, your terminology, your decisions, your internal debates — and it uses all of that context whenever anyone calls on it. It's not just a chatbot. It's an AI that's actively learning your company from the inside.

Anthropic are positioning this as the thing that makes Claude genuinely useful in enterprise settings, rather than a clever-but-forgetful assistant that needs everything explained from scratch every time. And to be fair, that's a real problem. Most AI tools at work feel like hiring someone brilliant who has amnesia. You spend half the time re-explaining context that the entire team already knows. Claude Tag is trying to fix that.

Here's the bit worth thinking about, though. The reason this is strategically interesting isn't the Slack integration itself. It's the data. If Claude is the thing that knows your company's institutional knowledge — the decisions made in 2024, the client relationships, the product debates, the internal thinking — then it becomes very hard to switch away from. Your organisation's context is effectively stored inside Anthropic's system. That's not necessarily sinister. But it is a significant commitment. And it raises fair questions about what happens to that data, who can access it, and what model updates mean for the institutional memory you've been building up.

Worth watching: whether this starts a bit of an arms race. Microsoft's Copilot is already embedded in Teams. Google's Gemini is in Workspace. Claude being in Slack is Anthropic planting a flag in enterprise territory — and the differentiator they're betting on is depth of context, not just raw capability. That's an interesting strategic choice. Might work. Might get acquired.

Meanwhile, something genuinely odd happened in Hollywood this week — and it's worth a couple of minutes, because the story says something interesting about where AI sits in the cultural conversation right now.

There's a film in the works. A biographical drama about Sam Altman — the CEO of OpenAI. Directed by Luca Guadagnino, who made Challengers and Call Me By Your Name, so not a no-name. The film is apparently called *Artificial*. And here's the thing: Netflix, A24, Focus Features, and Warner Bros. have all passed on distributing it. Major studios, quietly declining a film about one of the most talked-about people in technology, directed by a critically acclaimed filmmaker.

Now, nobody's officially saying why. But the read from inside the industry is fairly clear: nobody wants to upset OpenAI. OpenAI is increasingly embedded in the entertainment and media business — content deals, tools used in production pipelines, relationships with platforms. The studios are calculating that antagonising OpenAI, or even appearing neutral-but-uncomfortable about Sam Altman's story, is not worth the commercial risk.

Which is, if you think about it, quite a notable thing. A biographical film is a normal part of how culture processes influential people. Steve Jobs got one. Mark Zuckerberg got one — famously. The fact that major distributors are reportedly stepping back from *this* one suggests that OpenAI's influence over the media and entertainment landscape is now large enough to create a kind of gravitational pull. Nobody has to threaten anyone. The incentives just... work.

It's worth noting: Neon and Mubi are still apparently interested, so the film may well get distributed somewhere. But the headline — that the larger studios blinked — tells you something about the power dynamics here that a thousand analyst reports wouldn't.

And then there's this. A story that's more uncomfortable than either of those. Meta — Facebook's parent company — ran an internal programme to collect employees' keystroke data. The idea, apparently, was to use that data to train AI models. Now, the programme itself had already raised eyebrows internally when it was announced. Employees had flagged concerns. But this week, it emerged that the data collected — including keystrokes from individual employees — had been accidentally made accessible to other employees inside Meta. People could see each other's data. That's a fairly significant breach of the trust that already wasn't especially high.

This connects to something that keeps coming up — and Harry wrote about it as an identity problem rather than a technical one: the pattern where organisations treat their workers as inputs to AI development without giving much thought to what that does to the relationship. You can't simultaneously ask people to trust AI tools at work, and quietly collect their keystrokes to build those tools without clear consent. The two things work against each other. And when something goes wrong with the data — as happened here — it doesn't just create a security problem. It confirms exactly what employees were already worried about.

Meta said it was an accidental exposure, not a deliberate one. Which I'm willing to believe. But "we accidentally let colleagues read your keystrokes" is not a reassuring sentence regardless of intent. If you're in a leadership or people-management role anywhere, this one is worth paying attention to. The question of how you collect, use, and protect employee data as AI adoption scales is going to matter more, not less, over the next couple of years.

That's your lot for today. Claude's learning your Slack, Hollywood's scared of Sam Altman, and Meta had a bit of an oops with the keystroke data. All very normal Tuesday stuff. I'll be back on Wednesday. See you then.