Synthetic Minds | Frontier AI Went Open, And It Cannot Be Closed Again

Synthetic Minds | Frontier AI Went Open, And It Cannot Be Closed Again

The Synthetic Minds newsletter offers short daily insights to get you thinking. If you enjoy it, please forward. All signals are powered by Futurwise. If you need more insights, subscribe to Futurwise and get 25% off for the first three months!

I have just launched the Intelligence Age Scorecard! It will help you understand how ready your organization is for the Intelligence Age.

Today’s topic: AI & Automation

The Capability You Want to Contain Already Shipped

A lab with no investors, funded by the public, plans to give a billion people a yearly full-body scan. A model anyone can download for free out-hacked a leading closed AI at finding software flaws. One force is behind both.

The lock on frontier AI is breaking. The most capable systems are going open, free and community-funded, and the same key opens the medicine cabinet and the armory.

Midjourney, the image-generator turned community-backed research lab, unveiled a 60-second whole-body scanner it wants in 50,000 locations, "intended for everyone," funded by the public rather than investors.

Z.ai released GLM-5.2 under an MIT license, a frontier-grade model anyone can download, run on their own hardware, and modify, at roughly a sixth of a closed model's cost.

Security researchers at Semgrep ran it against a real vulnerability benchmark: given only a prompt, the open model beat a closed frontier coding agent at finding access-control flaws, for about 17 cents per bug.

It is not alone. DeepSeek, MiniMax and others ship the same way, mostly from Chinese labs, matching closed systems for a fraction of the price.

Governments answered by tightening export controls on the closed models, the layer that is no longer the leak.

That's the open-source story. Here is the signal.

For a decade the safety plan was a chokepoint. Keep the best models inside a few labs, behind a price and a login, and let one export license decide who else gets them. That plan assumed capability stays scarce.

It no longer does.

An MIT license is not a product launch you can walk back. Once downloaded, the model runs offline, on private hardware, and anyone can fine-tune it. You cannot recall, embargo, or unship it.

So the export controls around closed models guard a door while the wall stands open. The capability they ration already sits on hard drives in a hundred countries.

And the cure and the weapon ride identical rails. The same openness that puts early disease detection in reach of a billion people lets anyone run a model that finds, or exploits, flaws in your software for pennies. You cannot democratize one without the other.

The defenders are not standing still. Some twenty companies, including Anthropic, Google, OpenAI and Microsoft among, have banded together under the Linux Foundation to find and patch open-source flaws before attackers reach them. But the same model that scans a codebase for the fix scans it for the way in, and by one member's count, AI has already surfaced thousands of live vulnerabilities with fewer than five percent patched.

When the players who concentrated frontier power are reduced to chasing it, the balance has already tipped. The argument that frontier power was concentrating in a few hands described half the world, this is the counter-current, and it runs stronger.

The last time a government tried to lock a general-purpose capability behind an export license, it was strong encryption, which Washington classed as a weapon, so protesters printed the banned code on a T-shirt and carried it abroad. Containment lost then. It loses faster here.

So the question you should debate is not whether to allow open models. It is sharper: when the capability you defend with is free, unrecallable and in your adversary's hands, what is left of your advantage besides the speed you use it?

The era of frontier AI as a scarce, licensed asset is closing. What replaces it is not a safer or more dangerous world, but a faster one, where the edge belongs to whoever moves first.

The Intelligence Age Scorecard

Frontier-grade AI has gone open: a free model anyone can download and run privately matches closed systems at a sixth of the cost, and the medical hardware around it is going community-funded too. This is a WAVE question — are you still watching this shift, or already verifying whether your strategy rests on an access advantage that has vanished? Benchmark where you sit:

Take the Intelligence Age Scorecard to benchmark your readiness for the next two quarters, and the next five years. Or read the public Intelligence Age Scorecard of IBM, Visa, Qantas, Woolworths, Telstra or Commonwealth Bank first.

If this newsletter was forwarded to you, you can sign up here.

Thank you.
Mark