Speaker 0: cloud fable five is officially back online, but uncle sam's fingerprints are all over it. if you think they're getting the same raw hypercapable model that dropped early june, think again. anthropic just slapped a massive digital governor on its engine to appease washington. so fable five is back out now after, massive export controls and whatnot. following an eighteen day global blackout triggered by an emergency us government export ban, ban, anthropic officially restored claude fable five to the public july 1, so yesterday. however, when the commerce department, the model was rereleased with aggressive new, safety classifiers that are heavily throttling its capabilities, sparking immediate backlash from developers. so this proves that getting a frontier model past federal review board means sacrificing its raw utility. the era of unhindered cutting edge ai deployments is over. developers are now paying premium prices for a model that operates on a political leash. what do you think about this, john? Speaker 1: yeah. i think, this is absolutely ridiculous. and, yeah, we've talked about this before about this us government be being well, in my opinion, it's dictatorship. they're also i think we talked about them asking, all the ai companies for just a percentage of their company just to give it to the government. so, yeah. so i've started looking elsewhere for, ai, llms to use, and i've come across, this one called secana. so fugu. so i'm gonna look into that one and see how that one, works. but, i'm glad table five is back. i haven't messed around with it yet. although i do find opus, 4.8 is decent, but, we'll have i'll have to see how throttled fable five is. i didn't get a huge chance to mess around with, with fable five too much, but, yeah, we can get really political about this. this is, but, Speaker 0: we can. Speaker 1: but let's i Speaker 0: think we just wanna take a step back on that and say, you know, we let's understand it a little bit. anthropic leading up to this point, you know, claim that mythos and fable and all these things were, like, they're gonna wipe the internet. we're gonna find all the security holes ever. it's like dropping a nuclear bomb. it was the biggest hype wagon ever from any startup. look how dangerous this thing is. and all of a sudden, someone said, well, if it's that dangerous, maybe we need to pay attention. and all of a sudden, everyone say dictatorships. like, okay. well well, are you a nuclear bomb for the internet, or are you not? which is it, mister, so anyway, they seem to be applying the same type of rules, with tattp, because 5.6 had to go through a full week of review, used by the government before, they were allowed to release. so i think anthropic was just the first. and we can say dictatorship all we want, but we're starting to see that chinese models are catching up or superior that are being released as open source, that are being trained on the models that we're releasing over here. so the problem is how do you control that leakage when you start to think that ai is a strategic advantage for a country? Speaker 1: i i i don't know. i don't think if it i don't think it's a strategic advantage. i think it's just the government trying to control. but that's my opinion. think, there's, like, there's i think there were two other models that have come out since the ban that, equal the, the capabilities of fable five and mythos. so, because of that, either the us is in their own little bubble and they're not paying attention to these other, models that are coming out and don't realize it, which i doubt highly doubt, or there's something else behind it, which, again, i don't wanna be conspiracy theorist. the thing that really irks me is that, like, it's just, i lost my train of thought. but, i just, it's Speaker 0: i so let me ask you a question. what's the difference between what the us is doing and what canada's doing with sovereign ai? Speaker 1: what's canada doing with sovereign ai? Speaker 0: we're we're creating our own ai that we're gonna be able to control. so our ai is not being, controlled by any other companies or countries. Speaker 1: i didn't know we were doing sovereign ai. i knew we were doing sovereign data. oh, yeah. Speaker 0: so mccartney announced that. i'm i'm just saying it's not i don't see it as being political. i see it as being a, you it's survival. you have to start controlling, the ai inside your country, the big ones, specifically. you know, how long ago? it was ago, year and a half ago. we started talking about all the ai controlled, drones, fighter jets, the air to air fighters, the ground to air, the air to ground, all the different stuff. and so at one point, you know, it's a military decision. Speaker 1: yeah. and we could cover this another time, but did you know that the us used groc to launch missiles at iran? over 2,000 missiles? Speaker 0: well, i don't know if that's true. i'd like to see it, but i bet they use, star shield. so Speaker 1: yeah. yeah. they did. it's from the independent well, we could talk about it next time. we'll put it in the thing, but, it's just it's, like Speaker 0: but that's the military contract star shield has. so if star shield happens to be using grok, okay. great. but, you know, it's a military contract. i'm i'm neither here nor there on it because, we've seen what china's doing with ai. we've seen what other foreign countries are doing with ai. i think, you know, Speaker 1: these other countries are releasing these other ai models that are just as capable. so what's the point in porting Speaker 0: the point is that anthropic, every time a chinese model comes out, complains that they, you know, basically trained off all their models and they have to tighten their models because every time they release it, someone else is just training their models on their models and stealing the weights and doing that stuff. so the lockdown be supposed to stop the leakage. i don't know if it ever will because most of these companies have, diversity as their strength, meaning their people are from all over the world. so how do you, really bet whether someone's gonna leak stuff or not? Speaker 1: there's there's no way. with the internet nowadays, even you can keep things as secure as you want. like, fibo five leaked even when it wasn't even released to other like, when it was just governments that were allowed to use it. so, like, there's no way that they're going well, maybe they'll find a way. but as of right now, there's no way that they can completely lock down any llm model. Speaker 0: so i agree, but they're starting to work on it. they're starting to try. that big move by, amazon basically they wanted to be the secure digital provider because they claim they have the hardware and infrastructure controls that, the us government needs to control, who has access to what. so they only want the public lms to run on their infrastructure. and that's why i think they blew the whistle on anthropic overall is they're all got in for these, military contracts that are worth billions and billions. Speaker 1: for sure. yeah. Speaker 0: so june 26, the government permits, highly restricted rollout of fable's sibling model, mythos five, to fewer than a 100, enterprise entrepreneurs. june 30, the commerce department officially lifts the broad export control. and, july 1, anthropic restores sable five globally across cloud ai, cloud cloud, aws. the consensus is the model feels significantly less powerful due to, overtuned guardrails. so the yeah. i think this is the story is that we're not gonna see exponential growth in capacity as a result of the guardrails that are being put on, by companies now, we're gonna see growth in a different type of way, i think. so that's where i think the story is. they're starting to limit these bigger names, and their biggest models as they start to compare. and anybody who's comparing at this high level will get this scrutiny. if they're not immediately scrutinized, they will be soon. and i won't be surprised if, we start to hear about bans of chinese models. Speaker 1: i wouldn't be surprised with that either. for sure. yeah. Speaker 0: you know, i use deepseek a lot. i use, there's a few other i don't know, Speaker 1: timmy. this new sukana fugu? Speaker 0: i have not, and there's a couple other models i'm really interested in trying. but, i've got client work and a lot of other stuff, so i can't, you know, play around too much. but, you know, we play around at, like, our builder cafe tomorrow where we're teaching people how to build ai agents from scratch, so that'll be a fun time. but anyway, let's get back to this. yeah. anthropic is hard coded real time safety classifiers targeting offensive cybersecurity methods, structural, biologic biology code, and latent extraction, which prevents the model from revealing its internal chain of thought. when these broad filters are tripped rather than failing gracefully, the system executes an automatic fallback scaffolding, silently handshaking with cloud environment to dump the context session to, claude opus 4.5. so if it starts crossing a line into topics, they don't want you talking about or they don't want you poking around about, they drop back to opus 4.8. Speaker 1: yeah. that's interesting. also, i don't know if anthropic's doing this, but openai i think it's anthropic too. they're gonna be asking for id to be able to access these models, of government id yet. so i think it's coming in july. Speaker 0: and it's not, surprising. Speaker 1: yeah. Speaker 0: alright. so fable five is finally free from government house arrest, but it returned to the wild wearing a massive digital ankle monitor that goes off the moment you ask it to do anything truly complex. is that heavily filtered, guardrail clogged fable five worth the premium enterprise token price? because, you know, it's $50 per million output tokens, or are you shifting your infrastructure to an unmonitored open source models? let us know what your deployment strategies are in the chat below.