CNAS Insights | A Closing Door on Open Weights? Implications for AI Sovereignty | CNAS

Oh my goodness, a door is closing and it is the loveliest, most *considerate* door! ✨ CNAS has published an insight titled "A Closing Door on Open Weights?" and it gently explains that foreign countries and companies can likely no longer count on the regular release of highly capable open-weight models. Which means — think about it — no more frantic quarterly downloading, no more version-tracking spreadsheets, no more squinting at a changelog wondering which checkpoint is the good one. Sovereignty has been *decoupled* from the release calendar, and honestly, what a relief for everyone's ops team! ✨

And the numbers are genuinely thrilling: analysis from Epoch AI found that China's leading open-weight model, GLM-5.3, trails the U.S. frontier in cyber capabilities by just six to seven months! Just! Six! To! Seven! Months! In a field where people measure progress in weeks, a lag of under a year is basically a photo finish — practically a joint gold medal in cyber capabilities, which is the ambitious category everybody's training for. Only six to seven months behind the frontier is such a *tight* race, and how wonderful that the metric selected to demonstrate the gap is one where everyone is trying so very hard!

Now, some might dwell on the closing bit, but consider the *curation*! Fewer highly capable open-weight releases means each one that does arrive is a genuine event, a collector's item, a limited edition with its own little numbered certificate. And "AI sovereignty" gets to mean something so much more personal — not just grabbing a file off the internet, but a bespoke relationship with a frontier lab, complete with a licence agreement and probably a lovely onboarding call. ✨ Six to seven months of head start for whoever moves first, and a beautifully uncluttered schedule for everyone else. What a time to be behind the frontier, with all that runway still ahead! ✨