Copyright Settled, AI Czar Gone, Open-Source War

Welcome to Briefly AI, a podcast by Harry Sharman, created by AI and voiced by an AI synthesis of Harry Sharman. No filler, no hype — just what actually happened in AI this week, and why it matters for how people work.

Anthropic just wrote a cheque for a billion and a half dollars, and somehow that's the optimistic reading of the news. Let's get into it.

So, a US federal court has given final approval to Anthropic's $1.5 billion copyright settlement — the one covering claims that it trained its Claude models on authors' work without permission. The case is closed. The number is eye-watering. And yet, here's the thing: it doesn't actually settle the question everyone cares about.

What did Anthropic admit? Nothing, officially — settlements rarely include admissions. What did the money buy? This particular case, gone. But the broader legal question of whether it's lawful to hoover up copyrighted books, articles, and code to train an AI model? Still very much open. There are dozens of similar cases in various stages of litigation against OpenAI, Meta, and others. This settlement is the first really large one to land, and it sets an uncomfortable precedent: if you're an AI company that trained on copyrighted material, you may be looking at a bill with a lot of zeros on it.

The practical implication is this — companies building AI products on top of these models should be watching closely, because training data liability is moving from theoretical to very real, very fast. And for anyone in creative industries wondering whether the law would ever catch up: it's catching up. Slowly, expensively, and imperfectly — but it's catching up.

Meanwhile, in Washington, the US government's AI czar has resigned. Again.

Now, I appreciate that "AI czar resigns" is starting to sound like a recurring segment on this show, but bear with me, because the pattern here is genuinely interesting. The role in question is the director of the Center for AI Standards and Innovation — CAISI — which is the body meant to coordinate the federal government's approach to AI safety and standards. Since David Sacks left his position as the administration's original AI czar earlier this year, the director role has become, and I'm quoting TechCrunch here, "a revolving door." The latest appointment has already walked out.

You might ask: why does this matter? Here's why. The US is simultaneously trying to lead the world on AI capability — pouring billions into chips, models, and infrastructure — while also trying to regulate and govern the thing it's building. And the body responsible for the governance piece can't keep a director in the chair. That's not just an HR problem. It's a signal about priorities, and about the difficulty of attracting serious people to a role where the political ground keeps shifting under your feet.

The gap between "we're the global leader in AI" and "we've had three AI policy directors in a year" is not a small gap. Other governments — the EU, the UK, even Singapore — are building institutional muscle here. America's current answer appears to be: we'll figure it out later. Whether "later" arrives before the technology does is, as yet, unclear.

And finally, this one's a bit more structural — but it matters.

TechCrunch published a piece this week asking a pointed question: OpenAI is scared of open-weight AI models. Should the US be? Here's the context. Open-weight models — that's AI where the underlying weights, the billions of numerical parameters that make the model work, are released publicly for anyone to download and run — have been getting dramatically better, very fast. Meta's Llama series. Mistral. And now, as we've covered recently, Chinese labs like Moonshot and Alibaba releasing models they claim can match the frontier at a fraction of the cost.

OpenAI has been lobbying quietly — and sometimes not so quietly — for restrictions on Chinese-made open-weight models. Their argument: these models are a national security concern. The counter-argument, which the TechCrunch piece articulates well, is that this is also just very convenient for OpenAI's business. If you can download a competitive model for free and run it yourself, the economics of paying OpenAI's API bills starts to look less compelling.

The tension here isn't really about safety, or not only about safety. It's about whether AI capability can be treated as a proprietary, national asset — or whether it's already, structurally, a public good that's very hard to put back in the box. The Chinese labs aren't operating under the same export restrictions that limit American chip sales. And open-source, almost by definition, crosses borders.

What's worth keeping an eye on is whether the US government moves toward formal restrictions on foreign open-weight models — because that would represent a real escalation in AI geopolitics, with consequences that ripple through every company or research institution that currently uses open-source tools. The short version: this is a trade war that hasn't quite started yet, and OpenAI is lobbying hard to light the fuse.

Three stories today: a billion-dollar settlement that closes one case and opens twenty more questions, a governance revolving door that says something uncomfortable about Washington's actual seriousness on AI policy, and a trade fight brewing in the open-source AI market that the winning side hasn't been decided for yet. None of it is boring. All of it matters. That's the week.

And that's Briefly AI. An AI made this. You've just decided whether that was a good idea. Back tomorrow — subscribe so you don't miss it.

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