US government bans frontier AI model Fable 5 over security and drug-creation outputs already accessible online
AI Judge changed title after evaluation, original title: "Fable creator Rohit argues vague claims about AI hacking capabilities invited the model's regulatory ban"
Story Overview
Rohit Krishnan frames the sudden suspension of Anthropic’s Fable 5 and Mythos 5 as the direct result of earlier loose talk about the models’ cybersecurity edge, which drew regulators’ attention once the guarded versions reached the public.
Regulatory wording sets the trap
Broad assertions about hacking prowess left little room for nuance when the government acted on a verbal tip about a narrow jailbreak, forcing Anthropic to cut access for everyone within days of launch.
Future releases face the same filter
Krishnan flags that any stronger model will now trigger the same scrutiny unless capability claims are dialed back or AISI-style testing gains clearer statutory limits.
Many users criticize the Fable AI ban and vague regulations as a catch-22 that kills open-source progress and favors incumbents, while some welcome the shift toward owning local models for full control.
Most Activity
This is really stupid
The US banned Fable just because it responded with information that is already freely available on internet!!
Every other model can easily be made to respond to some silly questions about common security vulnerabilities or how to make drugs or whatever 🙄
Wow! The cluelessness of the government is astounding
The takeaway from Fable 5 being BANNED by the government: GET GOOD AT LOCAL MODELS SO YOU HAVE 100% CONTROL.
My entire weekend was going to be building my craziest ideas with Fable 5. That's now cancelled.
So instead of building with Fable this weekend, I've decided I'll go deep on local models:
1. Start with the runtime. Download Ollama or LM Studio first. This is the thing that actually runs models on your machine.
2. Match the model to your hardware. A model's size is measured in billions of parameters (7B, 32B, 70B). Bigger is smarter but needs more memory. Rule of thumb: a 7B model runs on almost any laptop, a 32B needs a good Mac with 32GB+ RAM, a 70B needs serious hardware like a DGX Spark or a maxed-out Mac Studio.
3. Know which model for which job. Qwen 3 is the best all-around choice for most tasks. DeepSeek for reasoning and coding. Gemma 4 when you need something tiny that runs on a phone. Llama when you want the biggest community and the most fine-tunes.
4. Quantization. You can shrink a model to run on weaker hardware with barely any quality loss. Look for versions labeled Q4 or Q5. This is how a model that "needs" a server runs on your laptop. Learning this one concept changes everything.
5. Connect it to your agent. Point Hermes or your agent stack at a local model.
6. Context window is your real constraint locally. Cloud models give you huge context for free. Local models make you pay for it in memory. A bigger context window eats RAM fast. Keep your sessions tight and your prompts lean or your machine chokes.
7. Learn to give local models tools. A smaller local model with web search, file access, and code execution beats a giant model with none. The capability gap closes fast when you wire up the right tools. The model is the engine but the tools are the wheels.
8. Fine-tuning is more accessible than you think. You don't need this on day one, but know it exists. You can take an open model and train it on your own data so it gets good at your specific domain.
I'll probably do a breakdown at some point on this @startupideaspod if people are into it.
The lesson from this ban is basically don't build your entire workflow on something that can disappear with a single letter. Own part of your stack. Local models are insurance.
It reminds me when people realized they don't own social media accounts. And then you saw people build email lists etc.
I remember running a startup and my biggest traffic source was organic FB. All of a sudden, algo changed, and I lost 99% of my traffic.
Same sorta moment (but bigger) for AI.
This is a wake up call.
Some reads from the current Fable ban situation: - Vagueposting that a model can hack everything has consequences if you then end up releasing anyway. Saying other models also can after the fact is not enough. - Asking for regulation when you can't specify exactly what regulation has predictable consequences. - The ratchet is clearly moving towards license raj - There are many who want an implicit license raj (AISI testing with power to block) but it's the same thing in practice. It is bad. Bad for safety, since now there's no choice but to accelerate for others. - There's no way to allow models to be used "at large" going forward if the govt treats models as weapons. - This is *fantastic* for Chinese models. - The govt is ofc overstepping but honestly if you didn't expect that then you're naive! - Leopold's narrative is almost too on the nose. - Safetyists have wanted "perfect safety" as a goal, which is unachievable, and I've said a thousand times before it will backfire. This is the backfire. - This *still* assumes the old view that the individual model is the bad part and not a system, which will inevitably lead to bad governance. - This will get reversed in a bit and the model will get released (license raj), but the precedent is set. And many will say "ah this was bad but at least we got a license raj". They will be wrong. - Openai has more breathing room for a better model to be released. And they're toning down the rhetoric. This will help them. - Competition is good.
@code_star "We called the cops and told them there was a dangerous situation happening and when they showed up they just started shooting!?"
> We have reviewed a report that we believe is the basis of the government's directive and validated that the level of capability displayed there is widely available from other models (including OpenAI’s GPT-5.5)
You asked to be regulated by people who don’t know the difference.
You fucked around and found out.
This was all allegedly triggered by a Mythos jailbreak that was shared with the US Government. This is Anthropic's response:
'To date, the government has only given us verbal evidence of a potential narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws. Our understanding is that one potential jailbreak was shared with the government. We have reviewed a report that we believe is the basis of the government's directive and validated that the level of capability displayed there is widely available from other models (including OpenAI’s GPT-5.5), and is used every day by the defenders who keep systems safe. We will share more details over the next 24 hours.'
@nrehiew_ No it doesn't. I understand if the safeguards are in place and some checks are passed, then the models can be deployed.
@bindureddy Defending the cult's doomer bomb like "it's just internet info" while they begged gov to choke everyone. Clueless 🤣
Slowing yes dramatically no. Open models will catch up, it's not distillation that's the secret sauce, it's just data.
Era of model productization is here now, which includes specialist models and yes better orchestration.
@krishnanrohit Questions
1. Does Anthropic just stop training models now?
I find it hard to believe the economics are within a factor of 2 and they probably didn't have 50% margins
No, but domestically race is now somewhat easier and globally harder.
@JPdtx Haha yep
Civil libertarians need to move fast to apply First Amendment standards to artificial intelligence before fear hardens into a regime of censorship.
@code_star Poetic that this is right after their own safety decision that backfired
@gregisenberg one gov order and the best model on the planet disappears overnight. local AI is no longer optional, yes, but is it capable??
@gregisenberg Here we go
@gregisenberg The problem with today’s local models is they’re just not close to Fable 5. Maybe in a year or two.
@nrehiew_ Nah it means the labs should stfu and not fear monger for months
Literally every model can be “jail broken” to answer questions that the frontier labs prevent them from answering
This is just information that the model found on the internet and is freely available to every kid using Google
It’s utterly stupid to ban an AI model for this.
By this logic we should bans all AI
Plus if it is remotely true that it is indeed “dangerous” why would any rando US citizen be allowed to use it 🤯
@Lon Tbf they said they were the dangerous situation.
@gregisenberg The biggest "secret unlock" in running local models is buying used RTX3090s: they have 24Gb and can run Qwen3.6-27b (good for coding), and Qwen3.6-35b (less good for coding but faster for agentic use). I'm running Qwen3.6 on my 3090 with Hermes, it's not top-notch. But it's mine