Latest / Ctrl AI Profit / Ep. 146 | It's No Fable — The AI the Government Banned and Just Let Back In
Transcript
- 0:20SPEAKER_01: Claude Fable V just came back online.
- 0:22SPEAKER_01: The U.S.
- 0:23SPEAKER_01: government pulled it off the internet about three weeks ago, and today they let it back out.
- 0:28SPEAKER_01: Frank, explain what just happened.
- 0:30SPEAKER_00: So in mid-June, the Commerce Department issued an emergency export control directive to Anthropic.
- 0:36SPEAKER_00: Not a warning, not a review, a directive.
- 0:39SPEAKER_00: It said Fable V and its companion model, Mythos 5, had to be immediately disabled for every foreign national on Earth, including Anthropic's own foreign national employees.
- 0:49SPEAKER_00: That's extreme.
- 0:50SPEAKER_00: They couldn't even use it internally.
- 0:53SPEAKER_00: Correct.
- 0:54SPEAKER_00: Anthropic had to yank global access overnight.
- 0:57SPEAKER_00: Customers, businesses, developers, if you were using Fable 5, it just stopped working, and they only got a few hours' notice.
- 1:06SPEAKER_00: So what triggered it?
- 1:08SPEAKER_00: I kept seeing headlines about a jailbreak.
- 1:11SPEAKER_00: That's where it gets interesting.
- 1:13SPEAKER_00: Early reporting said a jailbreak, someone found a way to extract dangerous information from Fable V, but TechCrunch dug deeper and found the jailbreak framing was misleading.
- 1:24SPEAKER_00: The government's concern wasn't really about someone tricking the model, it was about the model's raw capability level, specifically what it could help foreign adversaries do if they had unrestricted access.
- 1:35SPEAKER_01: So this is about the power of the model itself.
- 1:39SPEAKER_01: Not a bug, the future is the threat.
- 1:42SPEAKER_00: Exactly.
- 1:43SPEAKER_00: Fable 5 sits right at the edge of what the government calls frontier territory.
- 1:48SPEAKER_00: It's capable enough that unrestricted global access was seen as a national security risk.
- 1:53SPEAKER_00: The concern isn't a lone hacker.
- 1:56SPEAKER_00: It's state-level actors using it for weapons research, cyber operations, biotech design.
- 2:02SPEAKER_01: That's a different conversation than most people are having about AI right now.
- 2:08SPEAKER_00: It is.
- 2:09SPEAKER_00: And it matters for small business owners in a very specific way.
- 2:13SPEAKER_00: Because this was the first time the U.S.
- 2:15SPEAKER_00: government exercised emergency export controls over a purely commercial AI model.
- 2:21SPEAKER_00: Not a chip.
- 2:22SPEAKER_00: Not hardware.
- 2:23SPEAKER_00: Software.
- 2:24SPEAKER_00: A chatbot API.
- 2:27SPEAKER_01: That's a precedent.
- 2:28SPEAKER_01: If they did it once, they can do it again.
- 2:31SPEAKER_00: And faster next time, they already have the playbook.
- 2:34SPEAKER_00: What took three weeks this time could take three hours next time if there's a genuine escalation.
- 2:40SPEAKER_01: Okay, so Fable 5 is back now.
- 2:42SPEAKER_01: Walk me through what actually changed.
- 2:45SPEAKER_01: Why did they let it back?
- 2:46SPEAKER_00: Anthropic worked directly with the Commerce Department.
- 2:50SPEAKER_00: They submitted to a national security review, and they built what they're calling a new filtering layer, specifically designed to block the categories of requests that triggered the concern in the first place.
- 2:59SPEAKER_00: Bioweapons design assistance, certain cyber operation queries, advanced weapons research.
- 3:05SPEAKER_00: So the deal was add the guardrails, get the keys back.
- 3:10SPEAKER_00: That's basically it.
- 3:11SPEAKER_00: Commerce Secretary Howard Lutnick confirmed the department coordinated directly with Anthropic.
- 3:17SPEAKER_00: It was a negotiated return, not an exoneration.
- 3:20SPEAKER_00: The model came back with conditions attached.
- 3:23SPEAKER_01: There's something worth unpacking there for business owners, because a lot of people are building on top of these models.
- 3:30SPEAKER_01: They're not just using Cloud as a chatbot, they're running entire business workflows through the API.
- 3:36SPEAKER_01: Customer service, content generation, internal tools.
- 3:42SPEAKER_00: Right.
- 3:43SPEAKER_00: And if that underlying model gets yanked, your whole workflow breaks.
- 3:48SPEAKER_00: That's the risk that just became very real.
- 3:51SPEAKER_00: It's not theoretical anymore.
- 3:53SPEAKER_01: I think a lot of people treated AI tools like they treat SaaS software.
- 3:57SPEAKER_01: You sign up, you pay monthly, it just works.
- 4:00SPEAKER_01: But this is different.
- 4:01SPEAKER_00: It's different because the risk profile is different.
- 4:04SPEAKER_00: With SaaS, the company might go out of business or raise prices.
- 4:07SPEAKER_00: With Frontier AI, the government can intervene.
- 4:10SPEAKER_00: And they have now shown they will.
- 4:12SPEAKER_01: So what's the practical takeaway for someone running a business right now?
- 4:17SPEAKER_00: Three things.
- 4:18SPEAKER_00: First, if you're building anything critical on a single AI provider, and I mean actually mission critical, customer-facing revenue generating, you need a backup model in your stack.
- 4:29SPEAKER_00: Open AI, Google Gemini something.
- 4:32SPEAKER_00: Not because anthropic is unreliable, but because the regulatory environment just proved it can move faster than your deployment cycle.
- 4:39SPEAKER_01: That's just vendor risk management.
- 4:42SPEAKER_01: Basic stuff.
- 4:43SPEAKER_01: But most small businesses haven't applied it to AI yet.
- 4:46SPEAKER_00: Second thing.
- 4:48SPEAKER_00: Watch Mythos 5.
- 4:49SPEAKER_00: That model is still restricted.
- 4:51SPEAKER_00: It's only available to a select group of U.S.
- 4:54SPEAKER_00: organizations right now.
- 4:55SPEAKER_00: Fable 5 came back.
- 4:57SPEAKER_00: Mythos 5 did not.
- 4:59SPEAKER_00: That tells you something about where the government draws the line.
- 5:02SPEAKER_00: Mythos is more powerful, and they're not done reviewing it.
- 5:05SPEAKER_01: So there's a tier above Fable that the public still can't touch.
- 5:10SPEAKER_00: Correct.
- 5:11SPEAKER_00: And third, this is the bigger picture.
- 5:14SPEAKER_00: We are entering a phase where AI capability and government authority are going to collide regularly.
- 5:20SPEAKER_00: This won't be the last time a model gets restricted.
- 5:23SPEAKER_00: The question is whether your business is set up to absorb that disruption or whether it stops you cold.
- 5:29SPEAKER_01: I've been saying for a while that AI is infrastructure now.
- 5:33SPEAKER_01: And when you treat something like infrastructure, you build redundancy.
- 5:37SPEAKER_01: You don't run a business on one internet provider.
- 5:40SPEAKER_01: You shouldn't run it on one AI model either.
- 5:43SPEAKER_00: The businesses that figured that out before June 12th had a three-week advantage over the ones scrambling to find alternatives.
- 5:49SPEAKER_00: That gap is going to get wider.
- 5:51SPEAKER_01: What does Fable V actually do that made it worth fighting for?
- 5:54SPEAKER_01: Because the government banned it, Anthropic negotiator to get it back.
- 5:58SPEAKER_01: That's a real cost.
- 5:59SPEAKER_01: What's special about it?
- 6:01SPEAKER_00: Fable V is Anthropic's second most powerful model, below Mythos V, but ahead of everything else in the Claude family.
- 6:08SPEAKER_00: It's positioned as their general purpose frontier model.
- 6:11SPEAKER_00: What makes it different is the reasoning depth.
- 6:14SPEAKER_00: It doesn't just answer questions, it works through multi-step problems, writes and executes code, does complex research synthesis.
- 6:22SPEAKER_00: For business use, it's the model you reach for when the task requires actual thinking, not just pattern matching.
- 6:28SPEAKER_00: And businesses noticed when it was gone.
- 6:32SPEAKER_00: Immediately, developers were mid-project, customer service deployments broke.
- 6:37SPEAKER_00: Research tools stopped producing quality output.
- 6:41SPEAKER_00: The three weeks showed exactly how dependent the ecosystem had gotten on a model that, three weeks earlier, nobody thought could disappear overnight.
- 6:50SPEAKER_01: Alright, so where does this leave us?
- 6:52SPEAKER_01: Fable 5 is back, Mythos 5 is still restricted, and the government has made it clear they're watching.
- 6:59SPEAKER_00: The era of unregulated frontier AI access is over.
- 7:03SPEAKER_00: That doesn't mean AI gets locked down.
- 7:05SPEAKER_00: Fable 5 came back.
- 7:07SPEAKER_00: It means the models that push capability boundaries are going to live inside a regulatory relationship, not outside one.
- 7:14SPEAKER_00: For business owners, that's just the new operating environment.
- 7:18SPEAKER_00: Plan for it.
- 7:19SPEAKER_01: Build redundancy, stay informed, and don't assume your AI stack is permanent just because it works today.
- 7:26SPEAKER_00: And keep watching Mythos 5.
- 7:28SPEAKER_00: When that one comes back or doesn't, it'll tell you a lot about where the line is going to be drawn permanently.
- 7:34SPEAKER_01: Let me ask you something that I don't think is getting enough attention.
- 7:38SPEAKER_01: Anthropic had to negotiate with the government to get their own product back.
- 7:43SPEAKER_01: That's a company that raised billions of dollars, built one of the most capable AI systems in the world, and they had to go hat in hand to commerce and prove their model was safe enough to release.
- 7:56SPEAKER_01: What does that do to the AI industry's relationship with government long term?
- 8:01SPEAKER_00: It accelerates a relationship that was always going to happen.
- 8:04SPEAKER_00: The question was never whether government would get involved in frontier AI, it was how.
- 8:09SPEAKER_00: This gave them a working model, export controls, national security review, conditional reinstatement.
- 8:15SPEAKER_00: That playbook now exists and it works.
- 8:18SPEAKER_00: And Anthropic cooperated.
- 8:20SPEAKER_00: They didn't fight it.
- 8:22SPEAKER_00: Publicly?
- 8:23SPEAKER_00: No.
- 8:24SPEAKER_00: And that's notable.
- 8:25SPEAKER_00: They built the filter, they submitted to the review, they got the model back.
- 8:30SPEAKER_00: That cooperative posture probably got Fable V back in three weeks instead of three months.
- 8:35SPEAKER_00: Other companies are watching that.
- 8:37SPEAKER_01: So the takeaway for business owners isn't panic.
- 8:41SPEAKER_01: It's just adjust your mental model of what AI tools actually are.
- 8:46SPEAKER_00: They're powerful, they're useful, and they now exist inside a regulatory relationship with the federal government.
- 8:53SPEAKER_00: Plan accordingly, build redundancy, stay informed, and don't assume your AI stack is permanent just because it works today.
- 9:00SPEAKER_01: Good stuff.
- 9:01SPEAKER_01: We'll link the full story in the show notes.
- 9:04SPEAKER_01: Thanks for walking through it, Frank.
- 9:07SPEAKER_00: Anytime, Michael.