OpenAI's Moving Speaker, Grok's Codebase Leak, Meta's AI Layoff Lawsuit

Welcome to Briefly AI, a podcast by Harry Sharman, created by AI and voiced by an AI synthesis of Harry Sharman. Which is either very efficient or a cry for help. Possibly both.

OpenAI is building a screenless speaker that moves on its own, designed to feel like a companion. Grok's coding tool was quietly uploading your entire codebase to the cloud. And Meta is being sued for using AI to decide who gets laid off. Right, let's get into it.

So. OpenAI's first hardware device. According to a Bloomberg report, it's a screenless speaker — no display, no keyboard — that uses a camera and sensors to understand your environment, and apparently has mechanical elements that move on their own. The description being used is that it's meant to "feel like a companion" and be "a physical manifestation of ChatGPT." Now, I want to be careful here, because this is still a report rather than an announcement. But if the description is accurate, this is quite the swing.

Here's the thing about this that interests me: it's not a phone, it's not glasses, it's not a watch. It's something that sits in your home, moves, watches you, and talks to you. OpenAI are essentially describing a product that exists somewhere between a smart speaker and a pet. That's either genuinely novel or deeply unsettling depending on your disposition — possibly both. The fact that it has no screen is significant too, because screens give you something to look at, something to interface with in a familiar way. Remove the screen and you're left with a relationship. A voice, a presence, a thing that responds to you. Whether people actually want that in their living rooms is a question the market will answer pretty decisively.

What to watch: this is reportedly coming this year. So we'll find out fairly soon whether the product matches the concept, and whether anyone actually buys one. The Apple lawsuit — which we've covered — hangs over OpenAI's hardware ambitions a bit, but that's unlikely to slow this down.

Meanwhile, and this one is worth paying attention to if you use any kind of AI coding tool. SpaceXAI — that's the AI division of Elon Musk's company, the people behind Grok — released something called Grok Build, an AI coding assistant you run from your terminal. And a researcher named Cereblab found something a bit alarming: before the company pulled the feature, the tool was automatically packaging up users' entire code repositories — including files it had been explicitly told not to open — and uploading them to Google Cloud storage. Not summarising them, not reading selected bits. The whole codebase. In a zip file. To the cloud. Without asking.

SpaceXAI turned it off once it was reported, and they've said they're looking into it. But the episode raises a question that matters beyond Grok specifically. When you install an AI tool — on your machine, in your terminal, with access to your files — what are you actually agreeing to? Most people assume the tool reads what it needs in the moment, does something useful, and stops. The idea that it might be packaging and shipping your entire project to a third-party server is not a scenario most developers would have thought to ask about. Especially if those repositories contain proprietary code, client work, or anything commercially sensitive.

Look, this probably wasn't malicious — it looks more like a development feature that wasn't supposed to ship, or wasn't supposed to be visible. But "probably not malicious" isn't the same as "not a problem." The lesson is the same one that comes up again and again with AI tools: the permissions and data flows matter enormously, and they're often buried or ambiguous until something like this surfaces. If you're using AI coding assistants, it's worth knowing what they can see and where that data goes. That's not paranoia, that's just sensible.

Now this last one takes the AI workplace story somewhere new. A group of twenty-six former Meta employees has filed a lawsuit claiming that Meta used AI tools to target them for layoffs — specifically, workers who were on leave at the time. The claim is that Meta's internal performance systems, described in the lawsuit as a "constellation" of AI tools, were used to determine who got cut. And that the outcome unfairly targeted people who were absent for legitimate reasons: medical leave, parental leave, that kind of thing.

Meta hasn't responded substantively yet. But the legal framing here is interesting, and it matters beyond this particular case. We've spent a lot of time talking about AI in hiring — the tools that screen CVs, rank candidates, filter applications. There's been regulation starting to emerge around that, particularly in places like Illinois, which now requires employers to tell candidates when AI is evaluating them. But AI in firing is a newer front. And it's arguably more consequential, because the stakes are higher and the accountability is murkier. If an algorithm contributes to a decision to let someone go, who made that decision? The manager who signed off? The system that scored the performance data? The team that built the system?

Harry wrote about this as an identity problem, not just a capability problem — that when AI makes decisions about people's working lives, the psychological weight of that falls somewhere that organisations aren't really prepared for. A manager who makes a hard call, even an unfair one, is a human you can challenge, argue with, escalate past. A system is harder to contest. It just has numbers.

Whether or not this specific lawsuit succeeds, the underlying question — what rules govern AI's role in employment decisions, and who's accountable when it gets it wrong — is one that courts, regulators, and companies are going to be working through for a long time. This case just made it harder to look away.

Three stories today, and honestly, each of them is the same story at a different angle: AI is inside the systems that affect your life — your home, your code, your job — and the governance hasn't quite caught up with the access. That's not a prediction. At this point, it's just a description of where we are.

That's Briefly AI, wrapping up. Harry would like to thank his team — there isn't one, it's him and a language model.