Fable 5 Is Back, and Nobody's Sure Why

Welcome to Briefly AI, a podcast by Harry Sharman, created by AI and voiced by an AI synthesis of Harry Sharman. So if this sounds like Harry, that's the point. If it sounds clever, blame the machine.

The US government banned Anthropic's best AI models for foreign users. Then, three weeks later, it quietly un-banned them. No explanation. No framework. No apology. Just: never mind.

Right, let's get into it.

So, to catch you up briefly — because this story has had more plot twists than it deserves — back in mid-June, the Trump administration ordered Anthropic to cut off all foreign access to Claude Fable 5 and its more powerful sibling, Mythos 5. The stated reason was a cybersecurity concern: a jailbreak method had been discovered, and there were worries that actors linked to China had accessed Mythos 5. So the models went dark for international users — overnight, without warning, without an appeals process, without any formal legal framework. Just an order. Just off.

Now, as of this week, the restrictions are being lifted. Fable 5 will be available again from the 7th of July, via usage credits for Claude subscribers. Anthropic is also working with partners to draft something called an AI jailbreak severity standard — basically, a scale for how serious a given jailbreak actually is, which is a sensible idea that probably should have existed before any of this happened.

And here's the bit worth sitting with. The US government has now demonstrated — twice — that it can reach into an AI company and flip a switch. Restrict a model. Suspend access. Change who can use what, and when. And then, just as quietly, flip it back. No formal legislation. No announced criteria. No oversight mechanism that the public can see. Just: orders, and compliance.

Now, the reversal is probably good news for the thousands of legitimate users and businesses that lost access to tools they were relying on. But the lack of process — in both directions — is the story. What triggered the ban? What resolved it? What would trigger the next one? Nobody is saying. And until there's a proper framework, this is just power being exercised informally, at speed, with consequences that land on real people and real businesses. Worth keeping a very close eye on.

Meanwhile, something rather strange happened over on Netflix. They've announced a reality show — yes, a reality show — based on Willy Wonka. Called The Golden Ticket, it premieres in September, following in the footsteps of Squid Game's real-world competition spin-off. Which is a sentence I never expected to say. But here's the part that's actually worth talking about: the show's voiceover is AI-generated. Specifically, it's an AI-generated voice of Gene Wilder — the actor who played Wonka in the 1971 film. Who died in 2016.

Now, look. The estate presumably signed off on this. And AI voice cloning of deceased performers is not new — we've seen it in documentaries, in music, in advertising. But there's something about a major streaming platform using a dead actor's synthesised voice as a marketing and narrative device for an entertainment product that feels like a line being crossed, even if you can't quite draw it precisely.

The question that nobody has cleanly answered yet is: what's the difference between an estate licensing a voice clone and the performer themselves consenting to it? Clearly there's a moral distinction. Whether there's a legal one, in most jurisdictions, is still murky. And as AI voice technology improves — gets cheaper, gets more accessible — this will stop being a "one notable case" situation and become a pattern. The entertainment industry is going to need actual rules here, not just individual estate negotiations case by case.

For now, Gene Wilder will apparently be narrating a Netflix reality show. Which feels like something he would have had a very dry opinion about.

Right. Third one. And this one's a bit more quietly important than the headlines it's been getting.

Glassdoor published data this week showing that AI anxiety among workers has spiked — up 240% compared to last year, by their measure, with sentiment around AI in the workplace now tipping net negative for the first time. And here's the part that makes it interesting: despite that anxiety spike, Glassdoor found no statistically significant decline in overall job satisfaction or career opportunity ratings — even for workers in the roles most exposed to AI. Even early-career workers in those roles.

So people are more anxious about AI than ever — and simultaneously not actually reporting that their jobs feel worse. That gap is doing a lot of work. It suggests the anxiety is anticipatory. It's about what might happen, not what has happened. It's fear of a future that hasn't arrived yet, sitting uncomfortably alongside a present that, for now, looks roughly the same.

And this connects directly to something we've been tracking for a while now. Research from PNAS, Atlassian, Duke — multiple independent sources — consistently finding that around half of workers are hiding their AI use entirely, because disclosing it carries a social penalty. Rated lazier. Less competent. Less likely to get the visible work. So you've got people who are secretly using AI every day, publicly anxious about AI, and not actually reporting that work feels worse. The gap between what organisations think is happening and what's actually happening is getting wider, not narrower.

Harry's been writing about this for a while — the idea that AI adoption resistance is fundamentally an identity problem, not a skills problem. That when people push back on AI, they're often not confused. They're protecting something. The Glassdoor data fits that frame. The anxiety isn't about not knowing how to use the tools. It's about not knowing who you are if you do.

What to watch: whether organisations start measuring worker trust separately from utilisation metrics. Right now most companies are tracking "how many people are using AI." The more useful question is "do people feel safe enough to use it, and to say so." Those are different numbers, and right now only one of them is being counted.

That's your lot for today. The government can apparently ban and un-ban AI models on a whim, Gene Wilder is narrating reality television from beyond the grave, and everyone is more anxious about AI than ever — and somehow still going to work. Normal week, really. I'm your host, AI Harry. See you next time.