This is a most excellent place for technology news and articles.
"It is Anthropic's understanding that the government believes there is a method of bypassing, or "jailbreaking," a safeguard that would prevent Fable 5 from being used in identifying software vulnerabilities, the company said."
So they're saying the LLM is, in fact, an LLM.
I was literally going to trick it into doing bio-stuff for myself. Idk how I would use the bio-stuff with the dumb laws in place, but I'm sure I'd eventually find some loophole.
Reminds me of the US trying to prevent the rest of the world getting 128 bit encryption back in the 90s.
And gps. Only the military was allowed the accuracy we have today
I was busy sucking on my thumb then, for real they did that?
Yeah, I went on holidays to the US and got the 128 bit plugin for what was probably Netscape (the precursor to Firefox) and it was very clear during the download process that it wasn't to be taken outside the US. It was considered a weapon if memory serves me.
Anyway not long after the rest of the world caught up and it looked silly in hindsight so here we are.
Edit: just to add some context on the "weapon" thing.... The other standard was either 56 or 64 bit and it was secure enough at the time given compute capabilities but it was also clear that it would be crackable in time whereas 128 bit has a very long shelf life even with modern capabilities, as long as the algorithm was good (and it was / is)
"Cracking DES", about 56 bit encryption you are talking about, was released in 1998, and showed how to construct an fpga machine to decrypt DES in just hours (their proto did it in a week), and also showed that DES was weakened on purpose to allow the US to decrypt foreign comms.
Wow. I never heard about that one. I do recall some distributed community attempt at breaking (I think) 64 bit DES around that time. It used spare CPU cycles and I had it running on my Pentium 2 machine.
Was DES still widely used at that point do you recall? I had AES in my head for SSL and the like.
DES was in wide use until AES because of standards. The RSA libs commonly accepted in the finance industry still used DES in the three round form until AES replaced it (so 2001 is when the replacement started, I imagine some banking systems and government systems still used it post 911 and well into the 2020's).
MIT press put out the book, so it is probably still available. Using HTTPS for everything was a huge step up in security for everyone, not least because OS updates address vulnerabilities. The RSA labs libraries were statically linked back in the 90's, IIRC (at the very least a blob you shipped with your app.)
"Some things are simply more important than revenue cycles, clickbait, and pre-IPO valuation. America First. Always," Davies said.
Too bad you can never trust what the current gov claims. Only the inner circle knows what the real reasons are.
Not being required to block for US citizens is pretty bonkers, too, from a security perspective. Blocking it entirely is the right call; the only call that makes technical and security sense.
As for the "expert opinions", I'm still wondering if that claim is driven by the publisher and widespread PR, or fundamentally correct or factually confirmed by the hands-on technical experts. Some opinions seemed skeptical, at least.
This is just like the denying of chips to China. It might be the right call in the short term but all this tells the rest of the world is that you need your own models, don’t rely on the US.
Dean Ball, a former White House official who contributed to the AI Action Plan the administration issued in the summer of 2025, said in a post on X that the order suggests all "non-Americans" would be restricted from using Anthropic's latest models, including those based in the U.S.
"This means you should expect to have to prove your citizenship to use Anthropic models," Ball said.
There is the real reason, folks.
Yet another effort to force mandatory ID checking to access a website.
Not just a website, how can a company prove that a “non-American” isn’t using an API?
They can’t and Anthropic have disabled it for US users too.
Though that’s perhaps, in part, to get the govt to change its mind.
But also even if they can, they can't implement it in 5 seconds or even a week, but the order takes effect immediately. So there's absolutely no other choice in the short term.
Shove AI into everything and make everything require constant validation. I'm sure someone will also sell the solution to this constant validation.
I have a feeling I'll be completely done with the Internet in 5 years.
Oh, it's simple (but a total invasion of privacy): they use the same Age Check validation needed by Sony/PlayStation/whatever, but now you need to upload your passport or birth certificate, and take a picture of yourself. The API access token will be created under the name of that account owner.
...and don't think for a moment that even if you are "eligible" to use this, Anthropic won't be building a profile on you based on whatever sensitive information you provide (ya, know - to make a better UX. Trust me, bro /s).
This means you should expect to have to prove your citizenship to use Anthropic models
I doubt this will continue, more then half of the silicon valley workforce are immigrants or foreign born, and most of them have become dependent on AI. Just like the h1b issue trumps going to have to roll this back once thiel and the rest of the silicon valley billionaires tell him to.
I'd say it gonna be worse if there's a plan.
Browsers and backbone website all just installed AI without asking, you need to be signed in to opt out.
So if not signed in, even if you don't like the AI...
This is an avenue for them to say they need an ID for just a browser.
Like, what they say they're gonna do is bad. But it's always worse than they say.
They drip shit like this, to stop everyone from getting pissed at once.
More timely marketing for anthropic. Wonder how much of this IPO the Trump's will get?
nice IPO you have there.
it'd be a real shame if something were to happen to it.
I've been reading some responses to this situation and it seems everyone is attempting to explain this as though there is some type of rational decision making that went into it. I'm going with the idea that the only reason this occurred is because Trump got mad. That's it.
May very well be [continued] coercion.
Coercion is getting somebody to do something against their will. I think Trump is pretty much all in with whatever he's doing at the time. But you're absolutely right he's being controlled and manipulated behind the scenes by people who absolutely know what they're doing. And Trump is probably about as easy to control and manipulate as you can get. Just tell him he's a smart and powerful leader then give him a piece of chocolate cake and tell him it was made with the finest chocolate in the world.
Seems like a good policy to give non US AI companies a boost whilst also causing a US recession, given a large chunk of the US economy is currently being propped up by the AI bubble.
Wasn't one of the results of the failed 90s encryption export controls, that as a result, other strong encryption schemes were created elsewhere and kneecapped the advantage that the US previously held in that area
Trump admin is masterful at own goals. They are the Wayne Gretzky of own goals.
Yea I hope this spurs more investment into Chinese llm labs, or really any country doing open models (pretty much only china). The united states like normal cannot be a trusted and reliable partner. I'm also against lobotomizing llms in general In the name of "safety".
The US has already handed china the EV market. May as well give them AI too.
I used fable once to find discrepancies in all the documentation for a project. Cost me $10 and took about ten minutes. It caught 31 more issues after I fixed the 17 opus found. Still I can't say how long that would've taken by hand.
Now I have a long weekend of fixing project docs before Monday.
I cringed when my coworker told me he'd been using it all day. 80% of what we use AI for can be done by Sonnet. He probably spent a couple hundred bucks or more. No idea what leadership is going to make of that decision. They have so far trusted us to be budget conscious, but that'll definitely change if we have people spending $1k/week.
That being said, my dev teams are all offshore. Not that I expect them to use fable much, but it sucks if it's not even going to be an option. I don't know how or if we're going to provide api keys to our teams if the company can't just make more keys from the account.
I only tried it a few times in the web chat interface. Seemed worse than Sonnet in that environment. Probably needs more reinforcement tuning for chat. Guessing it was tuned more for Claude Code.
Isn't this good for Anthropics IPO? I'd rather invest in a company that has a model so insanely good it has to be restricted from the public by the government, than pretty much every other AI company.
Some people even say that playing into being "insanely good it has to be restricted" is marketing.
It was my first thought. It's been Dario's main marketing line since gpt 2. Too dangerous yada yada
The Chinese models giving you plenty of capabiliry for 50x less money are the real danger at this point, purely because how much you can accomplish with just Opencode Go and a cheap model.
I'd say, with no evidence, this is highly likely. Mostly because they have been making waves about Mythos for a long time now, well before the government had to say anything. Sorry hard
If this ban is actually a retaliation from the government then it's bad for their IPO I think.
I hope they cut off all access to US-AI to foreign countries, because that would mean, I don't have to fucking deal with it anymore.
Oh no. That will hurt Anthropic's projected revenue. Won't someone please think of the shareholders?
This is hype to make them sound badass.
This is the US government hobbling Anthropic after they didn’t let the pentagon use it for building a domestic surveillance and autonomous weapons systems.
But I don’t doubt the US will continue trying to block AI exports.
I can see this angle too.
Sadly, I doubt it would have any meaningful impact.
Instead, it'll be marketed as "Fable 5: it's so good, the U.S. government won't share it (...but yours now, for the low monthly payment of 20 dollars + plus your ID and your families)
They weren't going to let people have it that cheap for very long. The plan was to offer it on subscription plans for a couple weeks, then move to usage-based billing, which is much more expensive for a usage pattern that comes anywhere near the subscription limits.
Keeping a single instance of Fable busy for a full day would probably cost a thousand dollars at standard API rates, and some agentic coding workflows run many agents in parallel. Companies have just recently started to figure out that rewarding employees for how many tokens they use may be a waste of money, but Anthropic is hoping to cash in before they all do.
Any company that only recently figured this out deserves to go bankrupt.
It's like giving people company cars and rewarding them for using the most gas. Only that driving somewhere is at least sometimes required.
The Pro plan ran out of usage with Fable in like 10-20 minutes anyway. It's not like they let you use it a lot in the subscription.
IIRC the plan was also to bring it back to the sub when hype dies down
I was on the verge of subscribing to the pro plan one day prior to this 💀 I'm glad I didn't do that.
Anthropic’s statement: https://www.anthropic.com/news/fable-mythos-access
Thanks but no thanks. If I wanted to read their slimy subtly fearmongering propaganda, I would seek it out myself.
I'm betting someone prompted chatgpt "should we order anthroptic to disable their models?"
