Fable 5 Is Gone. GPT-5.6 Got Slowed. And Chinese Models Are Catching Up. Nobody Is Screaming.
Something uncomfortable happened this week. Most people missed it.
I posted something simple on X this week.
Not rage bait. Just an observation:
The same people dragging Dario Amodei for warning about dangerous AI capabilities went completely quiet when Chinese models started matching those exact capabilities.
Someone called it a dumb take. Muted me. Came back to call it dumb again. Then muted me a second time.
I’ll take that as confirmation I hit a nerve.
Here’s what actually happened.
June 9. Anthropic launches Fable 5. Most capable model they’ve ever released publicly.
June 12. US government issues export control directive. Anthropic has 90 minutes to shut it down globally.
Today. Fable 5 still not back for regular users. Mythos 5 only partially restored — approved US organisations defending critical infrastructure only. Everyone else still locked out.
Same week — multiple credible outlets including Bloomberg, Reuters and The Information reported that the Trump administration asked OpenAI to stagger the release of GPT-5.6. Limited preview first. Case by case approval for broader access. OpenAI and the White House haven’t officially confirmed it — but the reporting is consistent across sources.
So within the same month — America’s two most powerful AI labs both had their frontier models either blocked or deliberately slowed by their own government.
Been following this series since February? This is the chapter that changes the stakes. Subscribe if you haven’t.
Meanwhile — reports emerged this week from security researchers and Chinese AI circles that Zhipu AI, one of China’s top labs, is approaching or matching Mythos-level performance on security vulnerability detection. Not independently verified with public benchmarks yet. But being taken seriously enough that Western outlets started paying attention.
The exact capability class the US government used to justify shutting down Fable 5.
Reportedly showing up in a Chinese model.
No emergency meetings. No export controls. No national security crisis.
Just silence.
This is the thing I can’t get past.
When Dario warned about these capabilities months ago — fear-monger. Doomer. He invited the regulation.
When Chinese labs reportedly reach the same capabilities — interesting research. Competitive landscape. Nothing to see.
You can’t hold both positions. Either the capabilities are dangerous or they aren’t. Either everyone building them should face scrutiny or nobody should.
The outrage isn’t about the risk. It’s about who’s building the thing.
And while that argument plays out on X — serious builders are quietly moving toward open source. Because closed American frontier models now come with a risk nobody priced in: your government might take them away on a Friday evening with 90 minutes notice.
If this is making you think — restack it / share it. This conversation needs more people in it.
I don’t know if restricting Fable 5 was the right call.
What I do know is this — we’re winning the regulatory argument while potentially losing the deployment race. And the people who should be most worried about that are busy arguing about whether Dario is the villain.
He’s not the point. The capabilities are.
One question I’m leaving with:
Are we being strategically cautious — or are we tying our own hands while everyone else keeps running?
Drop it in the comments. I read every one.
— Alex, MBM
If you missed where this all started — February, a Twitter Space, a Pentagon blacklist: I’m Building AI Agents. This Week Made Me Question Everything. →
