Good morning, Fargo-Moorhead. I want to walk you through a story that is wild even by AI standards, because the clock on the useful part runs out this weekend.

Anthropic's most capable AI model yet, Claude Fable 5, launched June 9, got pulled offline June 12 under a federal export control order, and came back online July 1. Free access inside Pro, Max, and Team plans was supposed to end July 7, then Anthropic extended it to Sunday, July 12, after pushback from users. After that, using it costs real money, $10 per million input tokens and $50 per million output tokens, double what Opus 4.8 costs. If you have got a paid Claude plan, this weekend is your window to throw the hardest problem you have got at it before the free access disappears.

I have been using this model nonstop for the last ten days, and here is how we got here.

Where This Started: Mythos and the "Good Guys" Getting a Head Start

Back in April, Anthropic announced something called Project Glasswing. The idea was simple. They had built a new, unreleased model called Mythos Preview that had gotten frighteningly good at finding security holes in software. Good enough that Anthropic did not feel comfortable releasing it to the public yet.

So instead, they gave early access to a restricted group of the companies that run the internet's critical infrastructure. Cloudflare, Palo Alto Networks, Microsoft, and dozens of other organizations, around fifty in all, got to use it first, specifically to find and patch vulnerabilities in their own systems before anyone with bad intentions could get their hands on similar capability.

It worked. In the first month, partners found more than ten thousand serious vulnerabilities in software the whole world depends on, including operating systems, browsers, and open source code that powers half the internet. One bug in OpenBSD, a system built specifically to be hard to crack, had been sitting there undiscovered for 27 years.

The logic behind Glasswing stuck with me. Give the defenders a head start, because the offense is always going to catch up eventually. Good guys first, for once.

Then Fable 5 Showed Up for the Rest of Us

On June 9, Anthropic released the first version of this model family that regular people could actually use. Two versions came out together: Mythos 5 and Claude Fable 5, with Fable 5 built as the version for everyday subscribers. Same underlying model, but Fable 5 ships with extra safety guardrails layered on top, specifically around biology, cybersecurity, and AI research itself.

In plain terms, Fable 5 is the commercial, public facing version of the model that had only been in the hands of Fortune 500 security teams a few weeks earlier. For the first time, a small business owner with a Claude subscription had access to something close to what the big players got early.

Then the Federal Government Stepped In

Three days after launch, on June 12, the U.S. Department of Commerce ordered Anthropic to suspend both Fable 5 and Mythos 5, citing export control rules. The same capability that makes this model useful for defense also makes it useful for attack, and Anthropic did not have a way to instantly verify every user's citizenship or location at the scale a consumer product runs at. So they shut it down for everyone while they sorted it out with Washington.

It stayed dark for about three weeks. The export controls were lifted June 30, and Anthropic brought both models back online July 1. If you want Anthropic's own account of what happened, they posted it directly on their site.

The Free Window Closes This Weekend

Here is the part that matters most if you are actually using this thing. When Fable 5 came back on July 1, Anthropic said free access inside Pro, Max, and Team plans would run through July 7. After some backlash over how fast that cutoff was coming, they extended it to Sunday, July 12, at 11:59 PM Pacific time. For those of us in Fargo-Moorhead, that works out to just before 2 AM Monday morning, Central time.

One caveat worth knowing: the free access covers up to about half of your weekly usage limits, so it is not truly unlimited, but it is plenty to put the model through its paces.

After that, Fable 5 comes off the standard plan entirely. If you want to keep using it, you will need to turn on usage credits in your account settings and pay per token, same as the API: $10 per million input tokens, $50 per million output tokens. That is double what Opus 4.8 costs, and it is the most expensive rate Anthropic has ever set for a model available to the public. Anthropic says they want to bring it back as a standard subscription feature once they have the server capacity, but they have not given a date.

What I've Actually Built With It

I do not say this lightly after 30-plus years in IT. This is the first model that has made me stop and recheck whether I actually understood what "capable" meant.

I have been building nonstop for ten days. I finished an app for my networking group that I had been fighting with other AI models for weeks. Fable 5 built it right the first time. No circling back to fix bugs, no patching together half-working features. It just did what I asked, the way I asked it.

We also built a command center for the business itself, a single dashboard that ties together the tools we run every day, with Claude wired directly into it. It can see our email, pull from our connected software, and act like an actual digital assistant instead of a chat window you have to babysit. It does not always have a clean, plug-and-play connection to every tool out there, but the model is resourceful enough to find a way through anyway.

My wife Mona built two apps of her own. One is a health and wellness tracker for herself. The other is a book tracking app called Stackline, where she logs everything she has read, rates it, keeps a wishlist, and gets AI recommendations for what to read next. The feature that sold me: she can point her phone at an entire shelf at the thrift store, and it will scan every spine and tell her which books on the shelf match her wishlist or her taste.

The days of obvious AI slop, the stiff writing, the code that looks right until you actually run it, are over. This model builds clean, and it builds fast.

The Glasswing Lesson Applies to Small Business Too

Here is the thing about Project Glasswing that stuck with me the most. The whole premise was that the good guys, meaning big banks and cloud providers, get a head start on securing their own infrastructure before anyone else gets access to that level of capability.

Nobody is handing that head start to a small business in Fargo-Moorhead. Nobody is coming to audit your firewall rules or your email authentication before the bad guys figure out how to use AI against you. So I took the same idea and pointed Fable 5 at our own infrastructure and security instead of waiting for someone else to do it. What it found, and how fast it found it, was eye opening. If you run a business and have not had a real security review lately, this is exactly the kind of task worth testing this model against before the free window closes.

If You've Got a Paid Claude Plan, Use It Before Sunday

If you are already paying for Pro, Max, or Team, you have access to this right now at no extra cost, through Sunday night. My advice: do not waste it on easy stuff. Think of the hardest, most tangled problem sitting on your desk, the thing you have been putting off because it felt too complicated to hand to AI, and throw it at Fable 5. See what comes back.

Learn to Actually Use It: AI Classes This Friday

If you have heard all this and want real, hands-on guidance instead of figuring it out alone, our next AI class is this Friday, July 10, at Bonanzaville. We run Beginner, Intermediate, and Advanced tracks, and you can sign up at darkhorseit.ai/ai-classes.

Bottom Line

This is not going away, and after using this model for ten days straight, I am convinced we are heading into a real disruption in how things get built and done. The potential for an ordinary person to create real software, real websites, real digital tools, with no team and no budget, is genuinely here now. That is going to be great for a lot of people, and it is going to be hard on others. Nobody has a clear crystal ball on exactly how that shakes out. That uncertainty is exactly why we teach these classes, so that whatever shift is coming, you are ready for the speed and the quality these models can produce, instead of caught flat-footed by it.

If you want help figuring out where this fits in your business, or want a second set of eyes on your own security before someone else finds the gap first, reach out to DarkHorse IT. We talk about this stuff every Thursday morning at 7:40 AM on KFGO 790 AM.