Anthropic built an AI in February that was capable enough that they decided the world wasn't ready for it.
They call it Claude Mythos. It scored 97.6% on the US maths olympiad, against Opus 4.6's 42%. That's a different species of capability. And it's sitting in a vault, too powerful to be released to the general public.
But, I actually don't think that's the most important part of what's going on here.
Because also this week, Microsoft Research published a study of 388 employees, same tool (Copilot), same task. One group got standard training (ie. how to use the features of Copilot). The other got got thirty minutes on AI mindset and literacy training.
In the standard training group, 47% of them achieved what was judged to be top-quality output. And the group that got AI mindset and literacy training had 77% achieve top quality output. A thirty-point gap, from thirty minutes of AI literacy training.
What is AI literacy training, you might ask? It's understanding that AI is not a piece of software. It's an intelligence. And it responds to good communication techniques, to being given lots of detail and context, and to being delegated to.
So this week's video essay stitches those two stories together. I walk through why AI capability has stopped being your bottleneck (and why your people's AI literacy is), and what closing that literacy gap actually requires from senior leaders, as three practical starting points.
