AI regulation just arrived, and almost nobody voted for it. I watched a video that genuinely changed my mood about where this industry is heading, and I had to break it down for you. It comes from Matthew Berman, an AI commentator who usually radiates optimism, and even he sounds shaken this time.
Here’s the core of what the creator lays out. The US government reportedly asked OpenAI to stagger the release of its newest model, GPT-5.6. Instead of dropping it for everyone, it goes first to a small group of “trusted partners,” who test it for weeks before the rest of us get a turn. According to the original poster, this follows Anthropic pulling back Fable after access was restricted, and a similar limited-preview move with a model called Mythos.
Why is the video’s author so frustrated? He points the finger at fear-based marketing. He argues Anthropic and Dario spent months warning that frontier models are too dangerous for the public, then accused Chinese firms like Alibaba of “distillation hacking” Claude. The result, in his view, is regulatory capture: the biggest players get the best intelligence first, while startups, builders, and tinkerers wait.
🔑 The big idea: when only a handful of companies hold the frontier, power concentrates fast. Those labs keep using their best models to train their next best models, accelerating ahead while everyone else runs on six-month-old tech.
Here’s what stuck with me from his breakdown:
- ⚡ Competition cools off. The creator notes that rapid-fire model releases happened because OpenAI and Anthropic were fighting for every dollar. Slow that race, and you get fewer iterations, bigger jumps, and less real-world feedback. He cites Aaron Levy’s point that this looks like de facto AI regulation with no clear off-ramp.
- 🌍 China doesn’t slow down. The video’s author stresses that every delay matters in a global race. Pausing US access while rivals sprint, he argues, is the opposite of safety.
- 🛠️ Open source becomes the lifeline. This savvy professional says democratized, open-weight models are now more important than ever, and points to Bill Gurley and Peter Diamandis raising similar alarms about who gets to decide your level of access.
There’s also a sharp question the creator highlights from Gurley: if these labs are near superintelligence, why can’t their models detect espionage and distillation in real time instead of writing letters to Washington? It’s a fair jab.
A couple of practical takeaways the post’s author suggests:
- Try open source yourself. Download a model, run it locally, and learn the workflow. It won’t match GPT-5.6 yet, but support now is how it catches up.
- Contact your representatives. If staggered access bothers you, say so.
What I appreciate is that he’s clear OpenAI itself doesn’t seem happy about this. He quotes researcher Noam Brown hoping the model reaches everyone soon, and notes OpenAI’s own statement saying this kind of government access process shouldn’t become the long-term default.
I think this is a real turning point worth understanding, whether you agree with his framing or not. Watch the full video for the receipts, the quoted reactions, and the creator’s full argument. It’s the kind of discussion that’s better heard straight from the source.