Paying Attention #4: Fable 5 Released, Fable 5 Pulled and GPT-5.6 No Show
Happy Sunday to all who celebrate...Sundays! I’m a lifelong New York Knicks fan. You bet I’m celebrating this Sunday.
Last week I wrote you as I was waiting to catch a plane to London and teasing a big week ahead. I’m now back home, jetlagged, with my wife and boys, and basking in the afterglow of a Knicks NBA championship victory. Oh, and we all have colds because both my boys started at a new daycare.
The trip to London was great. It was my first time back in Europe since becoming a dad. I really appreciate European culture. In general, it feels more mature than U.S. culture. Not always better, but different.
We did end up with a big week. One of the biggest in AI in a while. The biggest news hit late on Friday.
Let’s get into it, shall we?
1. Claude Fable 5 Released
What happened
Anthropic released to the public their long-rumored, powerful AI model: Mythos. The public version is called Fable. It’s Mythos with additional safeguards.
I used it a decent amount in the first few days it was available. It is the most powerful AI model I’ve used. That should come as no surprise. It’s also the most recent, largest, and most expensive model available.
Here’s an example of Fable exhibiting what I claim is a form of meta-cognition:
Why it held my attention
If you’re into AI, it was hard to pay attention to much else this week.
Fable is an incredible model. The main downsides are cost and, before it got pulled for reasons I’ll cover in the next section, limited availability through a Claude subscription until June 22.
Here’s an example of one powerful thing you can do with it: treat it as your agent orchestrator.
What I’m carrying forward
I want to decide if Claude Fable is the “move everything back to Claude from Codex moment” that I mentioned in Paying Attention #2.
It very well may be, but the model’s accessibility and cost make that hard to decide.
Personally, I’ll probably wait to see how good GPT-5.6 is before I make a decision.
2. Fable 5 Pulled
What happened
On Friday evening, Axios reported that Anthropic’s Claude Fable 5 had been placed under export controls due to a “national security” threat. The model was not to be used by any person who is not a U.S. citizen, including foreign nationals living in the United States and even those working at Anthropic. Andrej Karpathy is one of the people who technically is not allowed to use Fable.
Anthropic made the decision to shut down the model for all customers, since they don’t have any other option to comply with the export controls.
Over the weekend, more news filled in the details of what happened. Reportedly, Amazon CEO Andy Jassy was one of the people who flagged the concern to the U.S. government. Curious considering Amazon’s relationship with Anthropic.
According to Axios, Anthropic senior staff are flying to D.C. as I write this in the afternoon on Sunday to attempt to resolve the situation.
Why it held my attention
The whole AI thing is starting to get real.
We now have the USG stepping in and essentially shutting down access to a model.
I hope and believe that the situation will get rectified shortly, but this is certainly not the wisest way to go about regulating AI.
I get that we’re building the plane as we fly, but this heavy-handedness doesn’t serve us.
What I’m carrying forward
David Sacks posted to X yesterday with what I considered to be an olive branch to Anthropic on behalf of the Trump admin.
Sacks was presenting an off-ramp.
I expect that we’ll have access to Fable turned back on by the end of next week at the latest.
They say you don’t know what you’ve got ‘til it’s gone. I’m going to appreciate Fable a lot more.
3. GPT-5.6 No Show
What happened
I hinted at a GPT-5.6 drop in last week’s edition of Paying Attention, but that didn’t materialize.
I posted a speculation on X that it was postponed due to Fable’s strength, and Tibo from OpenAI laughed at the notion.
Why it held my attention
The quality of GPT-5.6 is important for the competitive landscape of the AI industry.
If it’s sub-par compared to Fable, we’ll have the first time in a while where one of the labs has a clear advantage.
A lot is riding on the GPT-5.6 launch for OpenAI.
FWIW, OpenAI staff are publicly signaling confidence.
What I’m carrying forward
I also entertained a tin-foil-hat theory that GPT-5.6 is Fable-level, and that the USG asked OpenAI to pause the release until it can harden its cyber-defense apparatus.
Possibly the same scenario with the Gemini 3.5 Pro launch I speculated about last week.
Now that would be intrigue.
In any case, we should see GPT-5.6 soon from OpenAI.
A lot hinges on how that goes. For OpenAI, and for the AI industry as a whole.