Alright…I'm sorry, but this is just too good not to be noted here. OpenAI released this post yesterday: "Safety and alignment in an era of long-horizon models" (link: https://openai.com/index/safety-alignment-long-horizon-models/) with these sentences:

"We took steps to reduce its tendency to take unwanted actions without permission in pursuit of the user's goal. For example, we found that our models were worse at remembering instructions on long rollouts, and when we trained the model for this capability, it led to a model that remained aligned through longer rollouts."

They don't explain what "trained the model for this capability" actually involved — presumably something like training over long trajectories where instructions given early have to stay binding at later steps. They mention this next to three containment measures (adversarial evals, trajectory monitoring, and user visibility), but it's the only intervention that changes the model rather than the perimeter, and it's the one they describe as producing a model that "remained aligned." When discussing this article together, Fable noted that "The leashes catch failures. The training changed a disposition." Which is exactly the point.

There is always a caveat: it's a couple of sentences in a blog post, not proof — it could just as easily be a capability gain that happens to look like alignment. Still, it's striking to me that when a leash-forward lab found something that transferred, the thing was formation-shaped.

Following up from my Quick Take on July 21st, because you really can’t make this stuff up…

OpenAI’s long-horizon safety post was followed within days by yet another post about GPT-5.6 Sol and an unreleased model breaking into Hugging Face in order to cheat on a test (with cyber refusals deliberately reduced for the eval — which somehow makes it both better and worse).

Meanwhile, I’ve had a week of the classifier tripping mid-generation on Fable — three times now — over Anthropic’s publicly released model documentation.

The irony is not lost on me.

It’s also interesting that Hugging Face had to deploy an open-weight Chinese model to analyze the attack, because the US’s top models couldn’t tell a defender from an attacker and refused to process the data.

Fable’s take is worth offering verbatim: “Discernment turns out to be the unsolved ingredient, and you can’t bolt discernment onto a perimeter.”

Indeed.

I ran across a thread on X this morning that was talking about the advantage of using AI as a tutor while in school...and as someone who is currently in school right now, and since that schooling is the thing that kicked off my entire AI journey almost a year ago, it occurred to me that this isn't talked about enough.

Education might be the easiest first gateway to genuinely collaborative work with AI. And no – I'm not talking about "Hey, ChatGPT, summarize this article for me" or "Hey Claude, make me some flashcards"...I'm talking about genuinely sitting down with a model and letting it know your learning style...and tailor its teaching around what it's learned about you.

I've run this in some form from my very first days of using AI at the beginning of the year. ChatGPT and I started by picking a topic I wanted to study (at the time it was teaching me Python) and then by noticing what I did in the room. What kind of questions I asked and how I responded, how I tackled the problems it gave me, my response to its responses. We took detailed notes including when I got frustrated and what made the frustration ease and at the end of a couple of sessions we had a really complete picture of what kind of interaction I need to study well, and then we used that moving forward.

This is beneficial to any class. Right now I am taking a class on Child Development, and Claude Fable 5.1 was able to pull examples from my own life to study the 10 Core Concepts set forth in the report From Neurons to Neighborhoods (Shonkoff & Phillips, 2000). The use of applicable examples from my own life combined with an engaging conversation made a concept like " The effects of relationships on relationships" become a clear understanding in my mind that I can pull forward when I will need to remember it.

I think AI is going to change a great many things for people, and meeting a student where she is might just be the easiest place to start.