Two months ago I wrote: vendor-managed AI degrades silently. This week it did not degrade. It disappeared.
Monday: Anthropic launched the most capable AI model ever made generally available. Thursday: the US government ordered it pulled. Every customer, everywhere, same afternoon.
Seventy-two hours from “state of the art” to “currently unavailable.” In April I asked: “Is the AI still doing what it used to do?” The June update is simpler: “Is the AI still there?”
If your AI lives inside the product, on infrastructure you operate, the answer is always yes. It does not depend on a single vendor’s export-control status.
We built ours in-house for this reason. Some weeks the thesis proves itself quietly. This was not one of those weeks.