Project Glasswing and the Permission Wall

On March 29th, I wrote that AI was splitting into a two-tier intelligence system. Today Anthropic confirmed it.

I warned that institutions would get powerful AI and everyone else would get the budget version. I did not think it would happen this fast.

On April 10, Anthropic announced Project Glasswing. Here is what that announcement actually says.

Claude Mythos, their most capable model, goes to AWS, Google, Microsoft, JPMorganChase, and a hand-selected coalition of approved partners. Pricing after the preview period: $25 per million input tokens, $125 per million output tokens. No general availability planned.

Open-source maintainers, whose software runs most of the world's servers, received a $4M grant pool to share. The consortium partners received $100M in usage credits. That is a 25 to 1 ratio.

This is not a price wall. It is a permission wall. You cannot buy your way in. You have to be approved.

The organizations inside the coalition will use Mythos to find and fix vulnerabilities autonomously, vulnerabilities that survived decades of human review and millions of automated tests. The organizations outside will run the same unpatched software they had last week, while adversaries who want equivalent capabilities will find other ways to get them.

We already have a two-tier justice system, a two-tier healthcare system, a two-tier education system. Anthropic just made the tiering infrastructural.

Twelve days ago I argued that the deepest risk was behavioral, that cheap, sycophantic AI would degrade the epistemic habits of everyone who depends on it. Glasswing adds something harder: you cannot behavioral-standard your way out of a structural security gap.

The question is not whether this coalition does good work. It probably will. The question is whether the governance model it normalizes is the one we want.

Coalitions do not dissolve after the crisis passes. Permission systems do not voluntarily expand access. The architecture being built right now will outlast the safety justification for building it.

That is the fight. It is already underway.