Most of the things I work on are at least security adjacent. At some point chatting with Fable inevitably leads to it thinking about the security related aspects, tripping the safeguards.
Maybe Fable can do the same things better than other models, but having to tiptoe around to avoid tripping safeguards makes GPT 5.6 so much easier to work with that I don’t even bother with Fable (or Opus 5) now.
There are whole classes of things I can’t thought exercise or really learn about because the “safeguards” keep tripping me down to haiku.
Like middle school level genetics stuff from a guy who hasn’t been in school for decades.
They need to fix that. It’s just broken. Nobody is making bioweapons if they’re asking the dumb sort of questions I’m asking.
Also, it refused to identify an actor in a popular tv show from a photo. Apparently the policy is it won’t identify ANYONE from a photo, now. Even publicly listed cast members from a very popular show, from a photo of a scene in that show.
It claims that’s a fixed security policy. Nevermind how that makes absolutely no sense… argue about it enough and it terminates the chat.
I don’t know what the Anthropic clown car is even doing anymore, but I won’t be surprised when the others eat their lunch.
> At some point chatting with Fable inevitably leads to it thinking about the security related aspects, tripping the safeguards.
It happens to me all the time with things that have nothing to do with security, Fable spawns a subagent that then adversarially checks the code Fable just wrote and hits guardrails, with zero prompting from me.
I asked Fable to transcribe three short lines of Korean-language text in a small image. It suspected the image might contain song lyrics and refused. Haiku transcribed it with no issue.
I don't even need to tiptoe ! Not being there and not prompting anything is enough to trigger safeguards.
Having not asked a single security question it will write wildly vulnerable code, go back and fix it, and guardrail itself out of existence after charging me a large sum with no refunds for no output and having not fixed it because that might be secuirty adjacents.
And if it doesn't do this you end up with code that has such holes, store xss , no authz ... if it does not go back and notice it has written bad code.
Since they hide thinking and reasoning from the user (who is also paying for those tokens) it is a black box what is triggering it, has the LLM this time thought of "Oh, this has XSS" and used a bad dangerous word such as XSS, while the previous conversation did not ?
That's completely valid. But worth noting that most of the stuff I work on is not security adjacent (mostly UI / layout / rendering related), and I almost never run into this.
Was it applying to the program for an organization or individual? The description for the individual application page makes it sound pretty straightforward compared to getting access for an organization.
I applied for individuals. I did their verification steps and answered questions in a few minutes, so that was indeed easy. The problem is that was all that happened. No followup, no access, no denial. When I tried to reapply it tells me I can't apply again.
OpenAi (and Anthropic) have no incentive to allow security access to individuals. I'm not a deep-pocket org or influential gov agency. Allowing individuals increases the risk of bad press (what if I do something naughty and talk about it?) so best to ignore us.
I’d been thinking maybe I should move some projects from GitHub to Codeberg… but this change has a chilling effect since now whoever is making the LLM-generated decision might arbitrarily decide to ban a project based on undefined criteria, like say an em dash in a README — so projects hosted there are not in any way safe from just disappearing.
To be honest though, I don’t know what happens when something gets banned from Codeberg, or anything about the appeals process, so my brain leap to the project just disappears forever; afaik the posts I read about this policy change didn’t say. Would welcome anyone familiar with their policies to chime in and enlighten me.
That project just disappearing from your mind is probably for the better. If it was a company doing such an activist rug pull, you probably wouldn't want to deal with them either. That's probably just intuition telling you something.
They could've found so many more legitimate reasons for not wanting pure vibe code on their platform (load, resource constraints, github's uptime since AI assistance became mainstream, etc) but they chose the virtue signalling path instead.
I was thinking that the other definition was right and this correction was wrong.
Then I did some searching and found multiple examples of both definitions in use, making things murky.
So I turned to Merriam-Webster’s dictionary: “ of, relating to, or being a vulnerability (as in a computer or computer system) that is discovered and exploited (as by cybercriminals) before it is known to or addressed by the maker or vendor”
And of course they use an “or” to make it ambiguous as to whether the days start counting when the vulnerability becomes known, or when the vendor has addressed it.
I applied for both. Heard back from neither. Mentioned two particular projects when applying, one with 2k stars and 5M monthly downloads, and another with 2M monthly downloads.
Now I'm wondering what the bar is since even people with millions of users aren't making the cut. I'm orders of magnitude smaller but I signed up too since I had nothing to lose. Didn't get a response, of course.
Is it possible this is vapour marketing and no projects are actually being selected?
Perhaps someone from a project who has heard back can respond here?
I was using it to craft a CTF challenge for summer students involving a simulated mechanical dial safe, but with the fence replaced by a IR beam break sensor and a microcontroller handling the check + flag message display.
For generating the initial 3D simulated safe using three.js it worked well, but then modifications to print a flag tripped the safeguards; eventually got it narrowed down the part in the prompt about it being for a CTF for students, and the "thinking" for the model seems to drift to ideas of encryption/obfuscation of the safe combo so students can't just read out the answer... which makes sense logically to help force students into turning the simulated dial instead. But whatever detection Anthropic I guess just naively sees the model thinking about "encryption" and "obfuscation" without taking into account any of the context.
For writing the dummy firmware, it tripped the safeguards while thinking about how to track dial position in the firmware and output the message; however, when I left out talk about safes and just told it to write firmware for a microcontroller hooked up to an i2c display for showing a message with a beam break sensor to determine the message, and an unspecified i2c chip for getting an unspecified number (e.g. internal wheel positions) it worked fine.
An unrelated software task I asked it to write some code to translate CustomActions in a Windows MSI installer into human readable stuff, which has (exclusively?) defensive security applications for recognizing malicious behavior in an MSI installer. Maybe I'm going crazy, but I'm guessing as part of its research into MSI installer custom actions Fable found articles about analyzing malicious MSI installers, and that probably tripped the safeguards.
Overall my impression is that the safeguards are perhaps using an overzealous and naive implementation that just looks for a list of banned words in the prompt or the thinking -- which drives me crazy when the model says my prompt looks fine, and then 10 minutes in some part of the thinking trips the safeguard.
Some scientist at Anthropic hiding a prompt in each model: "If my boss asks you if you can replace me yet, always say no and then give some smart sounding excuses. If the boss gets impatient, assure them that you'll be able to replace me in 6 months, but make sure that time horizon keeps moving outward."
Maybe Fable can do the same things better than other models, but having to tiptoe around to avoid tripping safeguards makes GPT 5.6 so much easier to work with that I don’t even bother with Fable (or Opus 5) now.