Fable 5 apparently can't be used for coding? (This is from Anthropic's announcement.)
> After a series of productive conversations with the US government, we're redeploying the model with a new set of classifiers to target and block more cybersecurity tasks. In the near term, some routine tasks like coding and debugging will fall back to Opus 4.8.
> Fable 5 will be included for up to 50% of weekly usage limits through July 7, after which it will be available via usage credits.
> The new classifier also comes at the cost of flagging benign requests more often during routine coding and debugging tasks.
Here's Fable 5, the strongest model. Actually try to use it to harden your code and it turns into Opus 4.8. You have seven days to use it, and only half of that time's worth in actual usage. Enjoy.
Looks like it's going to be a thoroughly frustrating experience, even worse than initial rollout. For subscription users, the situation is almost indistinguishable from the export ban.
So fable will jump more often to Opus than it already did on original release? Working with fable felt like having to constantly fight against your work tool. Frustrating. Now they're making it even more frustrating.
Just a code review of my own project. Downgraded to Opus 50% of the time while evaluating the critical I/O and memory safety parts, the exact thing I wanted it to do.
I mean what do you expect when covering memory safety topics with a model that's not allowed to cover security topics? This seems totally expected. It'll be the same when 5.6 is released.
They're very similar models though, just with different safeguards and restrictions in placae around particular use cases.
I guess the underlying issue is that there is this model that is very capable, but it's being hobbled because of a fear of abuse. It may well be justified, but for a legitimate user any restriction just makes it a worse product and after all the puffery around how good it is (and some practical experience of how good it is) it's a pretty shit experience. "Here's our best model, no you can't really use it".
They're going to be verifying people's identities anyway. Why not put that bit of security theater to good use for once? I'm the author of project X, now let the model work on it, would you kindly?
This "only super special corporations get the model" nonsense is dividing society into haves and have-nots.
Fable is very strong for finding bugs. But you are explicitly not supposed to use it for cybersecurity. Even in the initial rollout I had it refuse and fall back to Opus when implementing a change password function
I can't roll my eyes hard enough at all the people who say this shit about Anthropic every day. I know I'll get downvoted. I know it's lame to complain about future downvotes. I don't care anymore.
Anthropic was correct in their assessment and early warning of Mythos's capabilities, and they did this rollout pretty well. They were not hype marketing. They were being genuinely cautious and honest.
The Trump admin was largely unreasonable with the sudden export control. (Though not entirely unreasonable.) The export control also had not much to do with Anthropic's pre-release warnings. See: GPT-5.6 currently being held up by the federal government.
> Anthropic was correct in their assessment and early warning of Mythos's capabilities, and they did this rollout pretty well. They were not hype marketing. They were being genuinely cautious and honest.
So what prevented them from putting in the sort of safeguards they ended up putting in without hyping it for months prior as being so good, it's too dangerous?
I'm not sure what you're saying. They spent ages adding guardrails to Mythos. Then they spent ages creating a whole new even more guardrailed version of Mythos called Fable. Then they added tons of classifiers so API requests to Fable would get rejected even if you ask a question like "what is a molecule". They put the thickest layer of bubble wrap around the model of any model in history. And then just today they made the classifiers even much more extreme than at the initial launch.
If they were truly honest in their beliefs of the potential risks of this model, how would their behavior have differed? I would expect exactly the behavior we see, if they were being honest in their belief.
Also note Dario here saying they shot themselves in the foot commercially with how they handled the rollout of the model - you can tell by his reflexive reaction how ridiculous he considers the accusation: https://youtu.be/v1wZwxY3CMg?t=2103
I am saying they could have not said anything about it being too dangerous etc. and just released Fable as a new model once the safeguards were in place and Mythos to trusted orgs as they did.
Instead they choose to hype for months about having a model that's simply 'too dangerous to release'.
In other words, why hype it beforehand instead of just quietly add the safeguards they ended up with anyways and release then?
Because Dario is an ASI doomer and deeply fears the power and danger of hyperintelligent AI and it'd be extremely irresponsible to not warn people "hey, this thing can do things the past models couldn't, and the future models may do it even better".
It was already frustrating to use before. I wanted to review my own code for OWASP top 10 kind of stuff and it kept refusing. It repeatedly popped up scary warnings about how I was violating TOS. I had to go through quite a few iterations of that prompt. When I finally got it to work it burned through all my remaining usage on a single run.
I won’t even bother with it if they’ve made it even more frustrating. Instead, I’ve been using a combo of Opus 4.8, GLM 5.2 and DeepSeek v4 Pro. Then I have Opus synthesize and verify the reports from all 3 and make the fixes.
This is what I’m thinking, too. OpenAI is gaining a structural advantage purely on the basis of not being considered an enemy of the administration. Anthropic really blew it with Washington.
They blew it by pretending to take some sort of moral high ground while their model was being used in Iran to blow up schools. I say they get what they deserve. I would have a lot more respect for them if they banned Pentagon use of their models outright.
i don’t think 5.6 will be as good as fable. their benchmark graphs say so, maybe they’ll take some limiters off next week or something now that being Fable tier isn’t scary anymore.
Fable 5 is switching into Opus 4.8 for everything I throw at it. It is not worth it Opus 4.8 is good enough for the next couple months. For $17/mo for pro anthropic is still worth it for claude code etc.
Yes, I am pretty sure it was simply poorly worded.
They almost definitely mean "you will notice even more false positives during seemingly routine coding/debugging tasks than you did at the initial launch". Which is not surprising, given the ordeal they've been put through. Hopefully it won't be too bad.
The main depressing thing for me is it's now only 7 days on the subscription, and then full API pricing, with no mention of even a plan to bring it back to the subscription in the future. (The initial launch mentioned two weeks of subscription, then API pricing, then a hope to return it back to the subscription not long after.)
I wonder if they meant to draw a link between cybersecurity coding and debugging specifically or this really will apply to all coding and debugging. If it really is a more general restriction, then this is practically the same as it still being restricted.
So, you can't use it for coding, can't use it for 'sensitive' information in chemistry/biology. It follows that it's likely bad for medicine and adjacent topics too.
What can you use it for? To run a breadth search on Erdos problems?
where did you find that? weird coz their post announcing this also mentioned Claude Code:
> Fable 5 will be available starting tomorrow, Wednesday, July 1, to users globally on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. For Pro, Max, Team, and select Enterprise plans,1 Fable 5 will be included for up to 50% of weekly usage limits through July 7, after which it will be available via usage credits. We will re-enable access on AWS, Google Cloud, and Microsoft Foundry as quickly as possible.
Reading the full blog post, I think the summary was just poorly written (because it's hard not to read that sentence like all coding is redirected to Opus).
> The new classifier also comes at the cost of flagging benign requests more often during routine coding and debugging tasks. As with all our safeguards, we’ll continue to refine this to better distinguish genuine misuse from legitimate requests and reduce false positives.
But wasn't the whole (claimed) reason that it got banned in the first place that it is a logical impossibility? Reviewing code for bugs is legitimate. Writing regression tests for bugs is legitimate. If the bug happens to be a security issue then the regression test may be a PoC or at least a step towards one.
Flowers for Algernon. And, sadly, expect this from now on. You saw it with OpenAI releasing Sol/Terra/Luna with a chart showing how they weren't quite as good as Mythos. It's all messaging to the USG to try to avoid/minimize arbitrary review from multiple agencies. 'Hey, it's smart, but look how stupid it is at "cyber."'
Holy moly this is upsetting to see on HN. If even here we're cheering on data center bans, AI is on track to become the next Concorde, or nuclear in the US. AI is the most amazing tech innovation that I've seen in my career since I started programming Perl back in 1994... Gosh, I'm gonna be gloomy for the next day.
A technologically impressive innovation that is ultimately doomed by being too loud and so expensive that it mostly benefits the rich before the costs just become too high for even that to be practical? That's the positive analogy?
We as a society should weight the cost/benefits of new technologies like this. What is the actual benefit of gigantic AI data centers? Is that worth the costs of the data centers to the power grid
If you're wealthy, it's not a problem because you probably won't have a data center next to you. If you're poor, you're screwed.
Those data centers require a ton of extra power infrastructure and the costs of those get front-loaded on the consumers already in the area, driving up their rates. The data centers get tax breaks because they can afford to buy the politicians, who get to claim progress and a bunch of other things that the poor won't see in their lifetimes, nor will their descendants. The progress and its outcomes might benefit society as a whole, in some small way, but the cost to society in terms of economic and environmental destruction will never be borne by the wealthy and will never equalize out because income disparity never lessens.
We're already starting to see some of the effect in lost jobs because business owners see AI as a replacement for technical labor. The people who are losing their jobs aren't being retrained and are becoming the equivalent of modern day coal-miners.
Meanwhile, their energy costs are rising to subsidize a data center that will be used to run an AI that will replace them and the owners will get richer.
But hey, at least the data center isn't in their backyard.
> Plus, turning off the ads should more clearly classify my blog as “non-commercial” for the various legal tests that impose greater liability on commercial actors.
Anyone know what these might be offhand? I think federal trademark law may sting more if used commercially. But what else could he be referring to?
Min release age sucks, but we’ve been here before. Email attachments used to just run wild too, then everyone added quarantine delays and file blocking and other frictions... and it eventually kinda/sorta worked. This does feel worse, though, with fewer chokepoints and execution as a natural part of the expectation.
Edit: bottom line is installs are gonna get SOOO much more complicated. You can already see the solution surface... Cooling periods, maintainer profiling, sandbox detonation, lockfile diffing, weird publish path checks. All adds up to one giant PITA for fast easy dev.
Min release age might just postpone vulnerability to be applied few days later in non trivial cases like this. More I think about it, Odin lang approach of no package manager makes senses. But, for that approach won't work for Javascript as it needs npm package even for trivial things. Even vendoring approach like golang won't work with Javascript with the amount of churn and dependencies.
Relatedly, I just registered for PACER to download court documents. It's pretty shocking that to get public legal documents the US Federal Court system requires full name, birthdate, address, phone, email, credit card info... and I THINK (it's past the initial registration page so can't confirm 100%) also mother's maiden name and 2 common security questions. Just a treasure-trove of PII if it ever falls into the wrong hands. (What's esp frustrating is even after going through this, I had to call a number and wait on hold for 1 hour to activate the account.)
It requires SSN or EIN -- almost all situations where you pay the government or they pay you require that as part of a law about enforcing federal debt collection.
Nowadays my writing (and maybe all of ours) has totally devolved into "prompt-ese." Much like days of yore where we all approached Google searches with acrobatic language knowing how to specifically get something done.
Now? I am pushing so much of my writing into prompts into AI where I know the AI will understand me even with lots of typos and run-on sentences... Is that a bad thing? A good thing? I am able to be so much more effective by sheer volume of words, and the precision and grammar is mostly irrelevant. But I am able to insert nuances and sidetracks that ARE passing vital context to AI but may be lost on people. Or at least pre-prompt-writing people.
Ok, but 3 generations ago, shorthand was a core skill that any competent professional could read and extract MORE value from than laboriously typeset prose. Something similar is probably happening now with prompt-ese and human-to-human (vs just AI) writing.
> Nowadays my writing [] has totally devolved into "prompt-ese."
I've noticed this myself. Even in my Obsidian vault, which only I read and write in. I think it's a development into writing more imperatively, instinctually. Thinking more in instructions and commands than the speaking and writing habits I've developed organically over my life. Or just "talking to the computer" in plain English, after having to convert my thoughts to code anytime I want to make it do something.
I've been thinking about the role of "director" in media as an analogy to writing with LLMs. I'm working right now on an "essay," that I'm not sure I'll share with anyone, even family (who is my first audience). Right now, under the Authorship section, I wrote "Conceived, directed, and edited by Qaadika. Drafted by Claude", with a few sentences noting that I take responsibility for the content, and that the arguments, structure, audience, and editorial judgments are mine.
I had a unique idea and started with a single sentence prompt, and kept going from there until I realized it should be an essay. So the ideas in it are mine. The thesis is mine. I'm going back and forth with the LLM section by section. Some prompts are a sentence. Some are eight paragraphs. I can read the output and see exactly what was mine and what the LLM added. But my readers won't. They'll just see "Author: Qaadika" and presume every single word was mine. Or they'll sniff out the LLM-ness and stop reading.
I can make a film and call myself director without ever being seen in it. Is is the same if I direct the composition of words without ever writing any of the prose myself? Presuming I've written enough in prompts that it's identifiably unique from cheaper prompts and "LLM, fill in the blank".
We credit Steven Spielberg with E.T. But he didn't write the screenplay. He probably had comments on it, though. He didn't operate the camera. But he probably told the operators where to put it. He didn't act in it. But he probably told the actors where to stand and where to move and how to be. He didn't write the music. But he probably had a sense of when and where to place it in the audio. And he didn't spend every moment in the cutting room, placing every frame just so.
But his name is at the top. He must have done something, even if I can't point to anything specific. The "Vibe" of the film is Spielberg, but it's also the result of hundreds of minds, most of whole aren't named until the end of the film, and probably never read by most viewers.
His contribution to the film was instructions. Do this, don't do that. Let's move this scene to here. This shot would be better from this angle. The musical swell should be on this shot; cut it longer to fit.
So where, exactly, is "Spielberg" in E.T.? What can we objective credit him with, aside from the finished product: E.T. the Extra-Terrestrial: Coming June 1982?
Uh, Steven Spielberg is all over E.T. For one thing, he storyboarded the big special effects sequences. He collaborated closely on the screenplay because it was drawn from his own childhood experiences. He was the final say in casting. His relationship with editor Michael Kahn is famously collaborative.
I think comparing your telling an LLM what to do and Steven Spielberg directing a movie just shows a total lack of understanding of how movies are made, and also inflates your own sense of your self.
> Uh, Steven Spielberg is all over E.T. For one thing, he storyboarded the big special effects sequences. He collaborated closely on the screenplay because it was drawn from his own childhood experiences. He was the final say in casting. His relationship with editor Michael Kahn is famously collaborative.
That's all meta. Trivia. Decisions he made or feedback he gave, that while influencing the final product cannot be observed in the final product (e.g. show me the actual Spielberg-drawn storyboard in the film; It doesn't exist, because the storyboard turned into a sequence of shots made by the cinematographer, instructing the camerawoman to point the camera at the actors lit by the gaffers, or into a work breakdown strucutre then followed by the SFX team painstakingly drawing it frame by frame). No one but Spielberg could say "That part was me, this part was Kahn's." I can't find any of that out just by watching the movie. When I engage with a piece of media, I presume the author is dead. What is in the media is canon, and what's not in it isn't. The behind the scenes, or the director's biography, or the interviews aren't part of the art. Art shouldn't rely on "Oh it's good, or even better than you thought it was once you know this cool fact or that wild story from production."
Star Wars isn't good only because George Lucas was a genius, or because they spent a lot of time on the models and tried a cool new text intro sequence, or because of any of the other novel effects. Lots of movies spend a lot of time in production, with a lot of experts and a lot of novel ideas, and still fail. Star Wars is good because the finished movie is good. We credit Star Wars generally as being George Lucas' brainchild, but if you know the backstory, it's only good because he had good editors to reign him in. But that's meta. Nobody knew that in 1977. They just knew they enjoyed the movie and it said "written and directed by George Lucas."
When I watch the movie I don't see the storyboard, or the redlines in the screenplay, or the casting notes, or the conversations and discussions with Kahn. All I know from the movie is the credits, and the credits don't say "Written by Melissa Mathison (with close collaboration by Spielberg based on his childhood experience)". Those are, from a lay viewer's POV, 'facts not in evidence.'
E.T. was a single example. I'm comfortable claiming my argument applies to all directors of all films, and all forms of art that are created by more than one person. Another example: "Over The Edge" and "Off The Wall", two books about deaths in US national parks. They each have two authors. Only one author co-wrote both of them. To whom do I credit my love for those books? Only to Ghiglieri, since I can see the consistent tone between them? That would be unfair to Myers and Farabee. Only to Myers and Farabee, because they're the park rangers that witnessed a number of the emergencies and deaths? That would be unfair to Ghiglieri. What about the editors, who surely worked hard to make books that are basically a list of stories about death interesting as a cohesive narrative. My only option is to credit all the authors, and everyone else involved, equally, and not try to break down paragraphs between "this author wrote this one, and that author wrote that one." They didn't distinguish, so I can't either. [1]
I'm all over my essay. I drafted and organized the original outline. I've made substantial changes to the order of paragraphs and what and how the arguments are built and developed based on my personal experiences. I am the final say for whose quotes are included and which ones are cut. My relationship with myself is famously collaborative (famous among my family and friends).
None of that matters to the reader. Whether I wrote it myself or with a friend, or used a ghostwriter, or used an LLM, the audience is going to credit or blame it on the name at the top. My papers in college weren't graded based on whether I spent 300 hours on them and revised them 20 times, or whether it was I or my classmate who coined that pithy line I then used throughout, or because I used niche knowledge about the subject I knew before taking the class. That's trivia. They were graded on the final single copy I submitted. I got once chance.
The only difference between an essay of mine being written by a ghostwriter I hired and an LLM is that the LLM output is always going to sound like an LLM. They are identical in that neither of them are "me". The ghostwriter will sound either like the ghostwriter or like the ghostwriter trying to write like me. But whether I hired a ghostwriter and published their work under my name, or if I used an LLM and the audience didn't notice, at the end of the day they'll credit or blame me entirely, because my name is at the top, no different as if I'd written the entire thing from scratch. I have no excuses except for the final product.
For this essay specifically, If I ever did release it or publish it, it would be under my real name. Firstly because I've never liked being "anonymous" online (I feel I never act or write like myself unless I'm speaking under my own name; opposite of most in my experience), and second because I would want the reader to know that there's a human they can credit or blame for it. I guess for me that's the tradeoff. When anonymous I won't use LLMs, because my ethos comes from being (and sounding) like a human being who merely doesn't want to share their name. Under my real name, however, I feel more comfortable saying "directed and edited by [real name], drafted by [llm]," because then the reader can decide if the ethos associated with my real name and affilations is strong enough to justify reading a logos and pathos that the human freely admits is not entirely from their own fleshy brain.
[1] They do, actually, at times. When one of the authors was directly involved in one of the stories and is recounting their personal experience, they will write "I (Myers)..." or "I (Farabee).." Aside from that they do not say who wrote what, or who influenced who.
That project is half dead now. There are commits, but has been no release in half a year, is missing major features (e.g. MCP server), and I haven't seen people talking about it for quite a while.
Nuts to let a 42k starred project just fade away. Seems like it could really have been something. I remember being supe rimpressed w it when I installed it in its first few months.
> routine tasks like coding and debugging will fall back to Opus 4.8.