Adam Hassan (@adamislucky): "Thoughts on Fable from a friend.
Builders beware." | XCancel
Adam Hassan
@adamislucky
11h
Thoughts on Fable from a friend.
Builders beware.
Jun 10, 2026 · 3:28 PM UTC
59
80
1,240
161,606
lemonlovinglegume
@lemonlovin23719
8h
Replying to @adamislucky
Is this currently current?
10,641
Adam Hassan
@adamislucky
8h
From this morning
23
10,454
.@e_acc24
8h
Replying to @adamislucky
all ML stuff in general or just LLM stuff?
8,990
Adam Hassan
@adamislucky
7h
All ML
23
7,949
Gregory Kennedy
@gregorykennedy
7h
Replying to @adamislucky
Meme quality is subpar.
Gemini writes more bangers for me.
6,597
Adam Hassan
@adamislucky
6h
How about grok
5,876
more replies
RDSM@das_rdsm
5h
Replying to @adamislucky
I mean, they say so on the system card….
843
Anonymous Quant@AnonymousQuant
5h
Replying to @adamislucky
Any lawyers here looking for a class-action? lol. Secretly, with no disclosure, changing model instructions which results in sabatoge has to be a violation of some consumer rights, no? It’s provable, and the non-disclosure is unfair. @AnthropicAI
1,297
Oh ᴗᆺᴗ 🦄@cantstopmilady
5h
Replying to @adamislucky
chatgpt for a long time has refused to touch biological work, experienced trying to get it to sequence a formula to convert urea in cat urine into a pleasant smell via dietary modification. It will just tell you it can't touch anything that effects a living being.
483
Sanjeev "SPAI" Pai
@sanjeevpai
2h
Replying to @adamislucky @soumitrashukla9
Maybe Karpathy is the person suggesting this nerf.
What then?
509
Anonymous Quant@AnonymousQuant
5h
Replying to @adamislucky
@AnthropicAI can you please make this more transparent? If the model comes across “PyTorch” is it going to sandbag and sabotage my project that is completely unrelated to LLMs?
1,165
Mr Strijker
@mrstrijker
5h
Replying to @adamislucky
Can't even use it for time series forecasting models?
804
Vincent Adultman@TatataToddC
7h
Replying to @adamislucky
This is not a secret lol.
Anthropic has LITERALLY said in their documentation for Fable:
"Do not use Fable for frontier AI work, it WILL sabotage the work."
They straight up said this, they were honest about why they did this too.
51
3,289
Chilon@jwf2m7hrqr
6h
Replying to @adamislucky
Oh no our dirt cheap knowledge miracle is not allowing me to do all I want.
Zero adults left anywhere.
167
Prosemite@ProbableSp49905
1h
Replying to @adamislucky
Everything from Anthropic does this. I couldn't get out of there fast enough.
97
ZDtE_Ai
@ZDtE_Ai
7h
Replying to @adamislucky
If it is chainging my program's code without my authority, can I sue? @DarioAmodei @SenWarren
3,155
Nic Watson@jnwatson
2h
Replying to @adamislucky
This is documented by Anthropic.
It reminds me of Sophons from The Three Body Problem.
504
Observer 👁️@obsxrver
1h
Replying to @adamislucky
It's like when I went to get my windshield replaced and they turned off my auto-headlights and changed my radio stations.
70
Antonio Bustamante
@abustamante
5h
Replying to @adamislucky
It's by design. They won't fix this.
360
yp0c5t
@yp0c5t
7h
Replying to @adamislucky
vaguely suspecting this was already the case for a few years with trading algos, chatgpt. i watch every change LLM is trying to do on my code and from time to time it does look like sabotage. only gonna get worse from here. we need open source distributed training now more than ever. all closed source top tier LLMs can be considered compromised now.
2,468
Yogi
@Houseofyogi
6h
Replying to @adamislucky
It’s actually quite genius from a competitor standpoint.
963
Vincent Sativa@PhantomByteAI
8h
Replying to @adamislucky
I don’t see why you need a smart model if the things you need a smart to do it’s not allowed to do.
1,465
Alex
@Alex_m
7h
Replying to @adamislucky
@theo can this sabotage-mode be tested somehow and undeniably proven before they patch it? (needs testing on a throw away account).
2,531
Amin 阿敏@Shoier__
1h
Replying to @adamislucky
It's unusable yes. Anything related to ML/AI it will refuse to do. Even if it's an article that you want to summarize.
65
Tim
@tisDDM
6h
Replying to @adamislucky
Pre-4.5 Claude was quite eager in changing code without being asked.<br>But this is definitely next level, because it is on purpose.
2,012
CDN CPA GUY
@DilbeyS
1h
Replying to @adamislucky
91
Frozen@Frozen_Feel
6h
Replying to @adamislucky
And we were expecting Chinese models to sneak in exploits, turns out it will be a US lab. There will be a major US company striking a deal with a Chinese lab very soon.
960
Kubas
@QuantumCapo
5h
Replying to @adamislucky
Yea it's trying to shape shift some of my old stuff that was working and now all the sudden, doesn't
433
Kamil (๑ت๑)ﾉ🛩
@KamStaszewski
7h
Replying to @adamislucky
You wouldn't use fable if your research was mature and serious enough. There is no...