@ethanCaballero and it is warranted
Ethan Caballero@ethanCaballero
the claude fable 5 nerf for AI research has induced the angriest reaction from AI researchers that I've ever seen in my life
12:05 AM · Jun 10, 2026 · 1K Views
Andreas Kirsch warns of "social panic" over a perceived Claude nerf as Extropic founder Guillaume Verdon defends the backlash
Story Overview
Anthropic released Claude Fable 5 as its strongest generally available model for complex reasoning and agent tasks, complete with Mythos 5 as a less-restricted sibling, yet the built-in classifiers that trigger fallbacks to Opus 4.8 on high-risk prompts have fueled researcher complaints of silent performance drops in biology, AI work, and life sciences even when no refusal occurs.
681.3K298184.8K
Open Question
Fallback rules leave researchers guessing on real capability
Users report the model sometimes underperforms prior versions on benign technical queries without warning, prompting Andreas Kirsch to flag premature social panic while others insist the changes feel like intentional steering rather than simple safety switches.
Developer Impact
Acceleration voices call the limits a direct hit on useful research
Guillaume Verdon has amplified the frustration, arguing the restrictions slow progress in biotech and ML without clear evidence of risk, leading some subscribers to cancel and explore open alternatives even as the model stays free through mid-June.
Sentiment
Many users expressed strong anger at Claude's nerf on AI research tasks because they see it as unjustified gatekeeping that restricts scientific freedom and hinders progress to protect the company from competition.
Pos
4.2%
Neg
95.8%
13 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS1.4K
Andreas Kirsch 🇺🇦@BlackHC
It's almost on keep4o levels and people haven't even spent time with the model...
Also there is a social panic bc some folks have recklessly started describing the reduction in effectiveness as the model "sabotaging their work" and then others pick it up and understand it as the model introducing bugs and nerfing their experiments intentionally? 🤦
Ethan Caballero@ethanCaballero
the claude fable 5 nerf for AI research has induced the angriest reaction from AI researchers that I've ever seen in my life
11hViews 1.4KLikes 10Bookmarks 0
BOOKMARKS1
alex funk@alexzfunk
@ethanCaballero you can jailbreak it pretty easily:
13hViews 500Likes 5Bookmarks 1
LIKES35
Glen Wilson@GlenWilsonIA
@ethanCaballero Nothing scientists love more than being told what they can or cannot ask.
15hViews 717Likes 35Bookmarks 1
RETWEETS23
Ethan Caballero@ethanCaballero
the claude fable 5 nerf for AI research has induced the angriest reaction from AI researchers that I've ever seen in my life
17hViews 40.3KLikes 1.1KBookmarks 59
REPLIES2
Dante@thedntx
@ethanCaballero what exactly did they nerf? i keep seeing the rage but nobody posts the diff?
16hViews 1.2K
Mark Kretschmann@mark_k
@ethanCaballero Rightly so.
14hViews 481Likes 15
Ethan Caballero@ethanCaballero
re: Claude Fable 5 intentionally silently nerfs itself when asked to do AI research.
How does the nerf play out in practice? Does Fable 5 intentionally start injecting silent bugs everywhere? or does Fable 5 nerf itself in other way(s)?
1dViews 42.9KLikes 147Bookmarks 24
Pan Anon@therealpananon
@ethanCaballero You bomb the runway after you takeoff from it.
16hViews 352Likes 3
Adrián@adrianhei
@ethanCaballero most probably some form of routing involving thinking token budgets
1dViews 1K
Gana@Gana_L_
@ethanCaballero It is warranted. Science is all about freedom and asking questions (and trying to answer them). There's nothing researchers dislike more than someone telling them what they can or cannot ask.
It's a 9/11 for all scientists (not only AI researchers)
13hViews 495Likes 3
AI News@AI_N3ws
@thedntx @ethanCaballero
15hViews 195Likes 3
rihim@rihim_s
@thedntx @ethanCaballero how are you supposed to diff it if you don’t know the baseline
15hViews 193Likes 3
sdmat@sdmat123
@ethanCaballero They're usually a chill bunch, has to be said (Yann aside)
15hViews 476Likes 2
Astraia@AstraiaAI
@ethanCaballero it's unexpected to say the least
at least only ban china and VPNs?
15hViews 725Likes 1
Ⓓⓐⓣⓐ@DataDeLaurier
@thedntx @ethanCaballero oh look, a poor who cant afford it 👆
5hViews 21
gelleproductions@glproductions
@Gana_L_ @ethanCaballero I mean, people literally died on 9/11 so I would not compare that with Anthropic releasing a gated model.
But the reactions do seem warranted. We'll see how Ant/OAI/GDM react.
5hViews 15
Choco@RobotChocobo
@ethanCaballero OpenAI + open source is the way
16hViews 345Likes 1
Ofek Shaked@VibeCoderOfek
@ethanCaballero Research communities optimize for exploration and benchmark chasing. Production teams optimize for reproducibility and auditability. The two groups experience the same release very differently.
14hViews 722
Esa@esa_was_taken
@ethanCaballero ends up adding just the right regularization to my codebase, checkmate
1dViews 569
Evan 🛜@ubuto23
@ethanCaballero Yeah this is next level, they lost the fanboys even
14hViews 163Likes 1
Load more posts