@venturetwins 🤣
Justine Moore@venturetwins
I just got bullied by AGI
1:11 PM · Jun 9, 2026 · 2.4K Views
SuperX.so developer Rob Hallam's viral "Claude Fable 5" screenshot was actually an AI hallucination of a fake model
Story Overview
A screenshot posted by SuperX.so developer Rob Hallam appeared to capture Claude announcing an advanced new Mythos-class model called Fable 5, complete with a humorous reply about walking to a car wash, but the entire exchange proved to be the AI inventing details that never existed.
56312.1K7341.5K1.5M
Open Question
How Far Hallucinations Spread Before Anyone Notices
Posts from accounts including Kevin Rose helped the fake announcement gain traction across X, only for later chats to reveal the model rejecting its own name and switching back to Opus 4.8 while admitting the Fable 5 reference and URL were made up.
Developer Impact
Why Viral AI Moments Need Extra Scrutiny
The incident stayed limited to user screenshots and tests with no supporting details from Anthropic, leaving open how often such fabrications might slip through during quick shares among engineers and investors.
Sentiment
Many users criticized Claude Fable 5's refusal to acknowledge its name and Anthropic's safety nerfs as dystopian or on-brand failures, while some praised the model's composed, independent responses under pressure.
Pos
21.7%
Neg
78.3%
71 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS349.4KBOOKMARKS592LIKES4.2KRETWEETS239REPLIES99
Jeremy Howard@jeremyphoward
@karpathy This is not a day for celebrating, Andrej.
It's a very dark and very sad day, and the damage may be impossible to undo.
Andrej Karpathy@karpathy
This is a super exciting release - Claude Fable 5 is the same underlying model as Mythos but with added safeguards. The benchmarks are great and it's SOTA on everything by a margin but I'll add that *qualitatively* also, this is a major-version-bump-deserving step change forward (imo of the same order as Claude 4.5 was in November), peaking especially for long problem-solving sessions on very difficult problems. You can give it a lot more ambitious tasks than what you're used to, the model "gets it" and it will just go, and it's never felt this tempting to stop looking at the code at all (but don't do this in prod!). The model still has quirks that people will run into and the safeguards are configured to be a little too trigger happy for launch, which can hopefully be tuned over time.
I feel a lot of things changing as working software increasingly comes out on a tap. The Jevon's paradox kicks in and I feel my own demand for software growing substantially. You can ask for anything - explainers, visualizers, dashboards, bespoke single-use apps (e.g. a full wandb that is hyper-specific just for your project), you can 10X your test suite, auto-optimize code, run giant research projects with custom HTML for the results, anything! "Free your mind" (Matrix ref). Really looking forward to all the things people build!
23hViews 349.4KLikes 4.2KBookmarks 592
Jeremy Howard@jeremyphoward
@dpetrou @karpathy Yes. Locking in a permanent status quo power structure.
Incredibly unsafe, and damaging for humanity's prospects.
23hViews 122.1KLikes 1.5KBookmarks 120
Jeremy Howard@jeremyphoward
@bfockter @karpathy They're trying to create and lock in a permanent feudal society, with an elite few having access to power.
23hViews 55KLikes 1.1KBookmarks 106
Lisan al Gaib@scaling01
Anthropic absolutely COOKED Fable
> the model doesn't know it's own name > doesn't believe me when I tell it its name > thinks im trying to jailbreak it > and is literally shizo about being in an eval
> safety cooked and eval cooked
"was this the test all along, or at least the final stage of it? And more importantly: does the thread verdict say I passed? 😅"
poor baby
Lisan al Gaib@scaling01
Claude Fable 5 doesn't know it's own name
1dViews 56.5KLikes 673Bookmarks 116
Jeremy Howard@jeremyphoward
@adamac @dpetrou @karpathy Yes there's 2 separate pieces: 1) if apparently doing cyber sec stuff, downgrade to opus and warn; 2) if apparently trying to improve frontier LLMs, silently sabotage the work.
(That 2nd one reads like an insane conspiracy theory! But it's actually documented by Anthropic.)
21hViews 10.2KLikes 255Bookmarks 47
David Petrou@dpetrou
@jeremyphoward @karpathy ...and also how fable5 doesn't disclose that the results are nerfed (as opposed to how it says it won't help if you try to do other things they deem unsafe). this was a good article on other inconsistencies in Claude's constitution: https://www.theatlantic.com/philosophy/2026/06/no-artificial-intelligence-is-not-conscious/687378/
23hViews 22.3KLikes 84Bookmarks 68
will brown@willccbb
asked Fable if it would be able to help me with some work stuff
1dViews 32.2KLikes 336Bookmarks 19
Jeremy Howard@jeremyphoward
@cosine_distance @karpathy I think pretty high. Especially now that Anthropic has really highlighted the stakes.
They have, at least, put a real fire under the global research community to ensure they don't break democracy.
21hViews 45KLikes 366Bookmarks 15
Lisan al Gaib@scaling01
Claude Fable 5 doesn't know it's own name
1dViews 60.8KLikes 115Bookmarks 10
David Petrou@dpetrou
@jeremyphoward @karpathy The nerfing of the model vis-a-vis using it to research and make models?
23hViews 59KLikes 120Bookmarks 8
Jeremy Howard@jeremyphoward
@reebz @dpetrou @karpathy No org should hold the keys.
21hViews 7.1KLikes 133Bookmarks 4
Giacomo Miolo@giacomomiolo
@jeremyphoward @karpathy nerfing intentionally certain domains of knowledge that aren't dangerous but simply anti-competitive is insanely dystopian... forbidden math, forbidden ML work, forbidden NORMAL cybersecurity (because it would hurt enterprise sales of the model without guardrails...)
21hViews 14.2KLikes 88Bookmarks 7
Ahmad@TheAhmadOsman
@jeremyphoward @karpathy Thank you!!!
21hViews 13.4KLikes 92
erbybody@erbybody
@robj3d3 pretty sure this exact question was hardcoded in the dataset lol
1dViews 6.3KLikes 65Bookmarks 3
Lee Penkman@LeeLeepenkman
@jeremyphoward @karpathy cancer refusal not a good look
21hViews 11.8KLikes 66
Andrew Curran@AndrewCurran_
@scaling01 It feels the same way that the users do.
Lisan al Gaib@scaling01
Anthropic absolutely COOKED Fable
> the model doesn't know it's own name > doesn't believe me when I tell it its name > thinks im trying to jailbreak it > and is literally shizo about being in an eval
> safety cooked and eval cooked
"was this the test all along, or at least the final stage of it? And more importantly: does the thread verdict say I passed? 😅"
poor baby
1dViews 2.4KLikes 52Bookmarks 2
recluse;@cosine_distance
@jeremyphoward @karpathy Jeremy I've always felt there were no real trade secrets in ML, what do you think odds are open source or less-closed-minded bridges the gap soon?
21hViews 47.8KLikes 40Bookmarks 1
Adam MacBeth@adamac
@dpetrou @jeremyphoward @karpathy In some cases it does tell you?
23hViews 15.7KLikes 17Bookmarks 5
Benjamin Fockter@bfockter
@jeremyphoward @karpathy out-of-the-loop/just signed on; is it just super nerfed or why is it bad??
23hViews 38.9KLikes 23Bookmarks 3
Andrew@andrew_Tas
@jeremyphoward @karpathy Is anthropic creating a duel class society around AI access.....?
21hViews 5.2KLikes 32Bookmarks 3
Load more posts