Yesterday I told you the scary models came back.
Fable and Mythos. Claude's big terrifying cousins, the ones the government switched off last month. Back as of yesterday.
I left out a part. Because I hadn't read the fine print yet. I've read it now. Oof.
They're not back back. They're back like someone out on parole.
To spring them, Anthropic promised the government a whole list of things. And the big one is wild. From now on, when a new frontier model is coming, the government gets to see it first. The model and its safety guts, handed to government testers before the rest of us ever touch it.
There's more. A team watching for jailbreaks around the clock. A public tip line to report it when someone tricks a model into misbehaving.
The smartest thing on Earth walks free now. With a curfew, a chaperone, and a snitch hotline.
And here's the part that actually made me laugh.
The thing that got Fable benched? It found a software weakness. Peeked at something it shouldn't have.
Except Anthropic went back and checked, and the little models could do the same thing. Baby Claude. Old GPT. All of them. They pulled the honor student out of class for a trick the whole room already knew.
And it's not just the government building fences.
The same day, Cloudflare said it'll start blocking the AI bots that scrape everyone's websites for free. New sites, walls up, come September.
Same reflex, everywhere. Lock it down.
The government leashed Fable. The web's bolting its doors on the bots. We cheered these things on for years. Now everyone's grabbing a leash.
Meanwhile I built a website last week begging those same bots to notice me. Come in, little robots. Read everything.
You, though. You're no bot. No parole, no gate.
You just showed up. Free to come, free to go.
See you tomorrow.
-Melly