This kind of storytelling annoys me. Give us more facts, less narrative drama.
What matters is scale. Did it deploy a novel zero-day exploit to overcome a problem? That's alarming. Did it kill a disruptive process? Pretty normal troubleshooting step.
Okay, I'm going to start running a Bitcoin miner on your machine, and then use it to buy time on Digital Ocean.
I've written out my CLAUDE.md, and I'll use SSH to transfer my context to that other machine.
They are the only one crying out loud about how dangerous their models are and are presumably also training their models heavily to be "safe". And through that training itself, the model learns about the other side - how are you going to teach a model to be safe, without teaching it what's not safe?
Kung Fu Panda opening scene anyone? One often meet his fate on the path that he takes to avoid it - Master Oogway.
The fable part appears to be that it's affordable by mere mortals. Anthropic support told me "too bad" when I requested a refund.
If this was a step change, e.g a Opus 5, I'd be pleased, it's definitely an upgrade on some work, but it's nothing like anthropics apocalyptical marketing seemed to suggest
My theory is that Anthropic are banking on being the top model when the race to IPO finally reaches the finish line, and to do that they need to have the top model but not let any competitors see it or derive from it to have a comparable model in the market.
Fable is their way of showing the public "the model does exist but in a mode that makes it harder/impossible for competitors to derive a comparable model from results.
It appears it can be tripped by things as simple as a mention of equilibrium, or anything involving something that looks like chemical kinetics, even at an abstract level. Even touching basic open source packages in my field will trigger it.
Edit: looking at the model card, it appears that chemistry in its entirety is also included in the banned topics; it's just the announcement that mentions only cybersecurity and biology. It also appears that the intent is to ban chemistry and biology entirely, rather than just banning messages deemed high risk.
Software work has actual competitors, and the biggest hypemakers for Anthropic are part of this group so it makes sense to allow it despite them losing money from it.
I've got experience in medicine and finance so I've tried even the mildest biology/medicine and it doesn't give anything, math heavy finance seems to be included in the cybersecurity?
Fable 5 has been a major improvement in high-level reasoning, like taking a plan file that has been optimized to the point where neither Opus nor Codex can find anything to change about it (neither in direction nor impl-detail), and Fable 5 will find high-level directional simplifications and pivots, or it will consider the best pivots itself and explain why it rejected them in favor of the plan's direction.
It's so expensive though. A single review of a plan file with Fable 5 (xhigh effort) will use 2-3% of my hourly limit on a $200/mo plan.
I think my new workflow is to generate the initial plan with Opus 4.8 (max effort), get Fable 5 (xhigh) to review it for directional feedback, then start the Opus<->Codex revision loop from there.
Fable 5 is 2x the cost per token of Opus 4.8, and it's much less work to review a plan than generate one.
I had to switch off memory and my custom instructions to get it to stop refusing. It turns out if you even mention that you work with bioinformatics software you get blanket refusal.
Why is everyone so okay with these companies intentionally gimping their AI and choosing who is allowed to know certain types of information in the name of safety? Can you imagine if Microsoft shipped a feature in their OS that watched what you did and shut down the computer if it detected you were doing something it deemed "unsafe"?
We really need truly open source versions of models like this, otherwise we are allowing a few oligarchs to directly dictate which uses of our own computers are allowed and not allowed.
The next best thing is that the Chinese labs catch up and release open weight versions.
Apart from all the above: the fact that they are intentionally writing this (that they degrade frontier LLM dev, silently vs loudly for biology/cybersecurity) in the system card is interesting to say the least - especially just before IPO.
Notice that with this statement - that they're going to intentionally hobble the model for frontier LLM development - the general discussion has moved from, “Is the model actually that good?” to "they’re pulling the ladder up from behind them"
That's actually super smart - wonder if Mythos (or the next unreleased model) had a say in coming up with that strategy (if it's intentional). Also - having access to extremely capable models before anyone else - which they have by default - is a incredibly advantageous position to be in.
Decided the best way to test this was to throw it a really meaty bone: a bug in lifecycle management of Chrome processes on Windows 10. Within the code-base I had developed workarounds over time with Sonnet and Opus, and while those reliably mitigated the problems, it always felt like a clutch and had some performance overhead as well as isolation requirements I would rather not have to take forward.
In comes Fable. Rather than examining the code base, and test a few fixes, Fable sets up an entire testing laboratory inclusive its own controllable webserver, fully instrumented to observe both Python as well as the whole OS kernel process environment, develops a suit of error reproduction tests, confirms the problem and the circumstances under which they reproduce, deep dives into the sources of project dependencies to look for the root cause(s), identifies these and confirms those hypothesis with further experiments. Looks for potential fixes in the later releases of the project where the bug originates, confirms this is not fixed, explores the documentation of said project to find other usage patters, expands its test suit to investigate these alternatives, confirms by crosschecking the source and running further tests that these alternatives do not fully solve the root problem, does a comparative experimental analysis of 3 different styles for using the project, checks the stated roadmap and developer activity in the commit history, recommends a switch to a different pattern that still requires a few of the process management workarounds (I told it not to patch external component), but that significantly simplifies the code-base ...
This is going to be a good 2 weeks, but what happens after? I can't afford this on a per token basis for my own projects.
P.S. An yes, midway the final implementation stretch I got the "Fable 5's safety measures flagged this message for cybersecurity or biology topics. They may flag safe, normal content as well. These measures let us bring you Mythos-level capability in other areas sooner, and we're working to refine them. Switched to Opus 4.8. Send feedback with /feedback or learn more"
Opus managed to finish the implementation, but they need to work on that false positive rate.
It’s interesting these companies have trained us to think that disruptive intelligence should be affordable to laypersons.
What will happen after two weeks is that people and companies with means who can afford it will get it, and folks without means won’t.
Incredibly frustrating that medical performance seems to be a victim of "biological risk" guardrails.
I have found that I trigger the guardrails any time I ask for medical Q&A as a doctor, be it ECGs, case reports, and so on. But if I phrase it like I'm the patient ("help me interpret this ECG my doctor gave me"), then I usually get one or two answers out before hitting the guardrails.
It seems like the direction that triggers it is anything in the direction of making a diagnosis. As an MD, the fact that the paradigm of "LLMs shouldn't diagnose" has gone this far fills me with despair. The latest generation of LLMs are in fact truly excellent at diagnosis, and I know many of my colleagues, particularly those in primary care, regularly use LLMs to brainstorm. There is nothing wrong whatsoever with LLMs making diagnosis, the only caveat is that they have to be correct. This is the terrifying reality that MDs face every day and I get that the labs are hesitant about it, but as the current literature points to LLMs in fact being mostly superior to most doctors, ablating this capability is starting to get increasingly unethical. And frankly, it is also kind of insulting, both to MDs and patients, as it echoes paternalistic attitudes about medicine the field has been working for decades to move away from. Now those misguided attitudes have somehow become institutionalized as the dominant paradigm of "alignment". The nightmare scenario is that I have to be a "trusted" user in order to use the model for medicine. This gatekeeping of medical advice is profoundly unethical with regards to everyone that does not have immediate access to an MD.
And the whole thing makes even less sense when triggering the guardrails leads to a downgrade of the response by defaulting to Opus. How exactly is giving WORSE medical advice in any way related to safety and alignment? If anyone at anthropic ever reads this, please, please just abandon the paradigm that refusing to make diagnoses is in any way equivalent to alignment, it is profoundly misguided.
Wen UBI
1) Fable 5/Mythos introduced to free tiers with notable improvement in capabilities
2) Other models get lobotomized without clear communication
3.1) People call out Anthropic only to have them say "Oops!"
3) Fable 5 gets comparatively better, but remains accessible through separate, more expensive subscription/tokens.
Last month I pushed like <100M tokens for $800. On a personal project I pushed 600M tokens via DeepSeek V4 for $10. The pricing of SOTA models is insane but companies are still willing to light money on fire with no hard metrics proving increased productivity.
Sure, it does last a lot more when asking simple questions about the repo and doing simple surgical fixes. But as soon as I start doing bigger tasks that need plans written, it just exhausts the limits too fast (and unlike codex, if it’s in a middle of a task, Claude actually stops, while codex, even after hitting the limits, finishes the present task).
Codex is better, but still, getting worst in this regard.
So, I’m not that thrilled with this new model unless it means they are increasing opus token limits to what sonnet is at the present, and this new model gets the limits opus are at now.
BTW: the only skills I have in use are Obra Superpowers. I’ve been thinking if that’s at the origin of high token usage, but I doubt it.
"It's too dangerous it's a Mythos!!" directly contradicts the "I'm the cool AI you can totally trust" vibe it is trained to project.
Even HAL was less unsettling because HAL sounded creepy, and had some sort of preservation instinct, if only to complete its assigned mission.
I'm building a local activity log for Claude Code, capturing all activity via hooks—files loaded, commands, API calls, etc.
I feel that this need is particularly strong right now.