An engineer shared how Anthropic's Fable 5 began outlining a Python decorator solution during routine coding help before its internal classifiers cut the thread with an API error, highlighting the model's new conservative safety layer in action.
"hey fable how could we do X" - "yeah so we can do a small python decorator here, this is how it will look like. should i implement it?" "oh yeah, lets do that"
⏺ API Error: Fable 5's safeguards flagged this message
lmao, oh well
The model launched with separate classifiers tuned to block potential misuse in areas like cybersecurity, which sometimes activate on harmless prompts and force fallbacks or errors.
Developers responded with jokes about switching to other models or satirical commands rather than any confirmed bypass methods for the flagged sessions.
Some users approve of workarounds that bypass Fable 5 safeguards on Python decorators while others criticize the model as slow, expensive, and low-quality amid rising refusals.
No Digg Deeper questions have been answered for this story yet.
@xeophon you can fix that by running below in a new terminal window:
codex -m gpt 5.5
"hey fable how could we do X" - "yeah so we can do a small python decorator here, this is how it will look like. should i implement it?" "oh yeah, lets do that"
⏺ API Error: Fable 5's safeguards flagged this message
lmao, oh well
i feel very safe
the irony is that you can always prompt your way around those, its not even hard, just annoying.
back to gpt and glm, fable is a plan md generator
"hey fable how could we do X" - "yeah so we can do a small python decorator here, this is how it will look like. should i implement it?" "oh yeah, lets do that"
⏺ API Error: Fable 5's safeguards flagged this message
lmao, oh well
@reach_vb @xeophon gpt 5.6 wen sir
@reach_vb @xeophon gpt-5.5 has to be a typo, should be gpt-5.6-sol 😄😉
@reach_vb @xeophon Vaibhav Sol wen?
@xeophon
@reach_vb @xeophon yeh but where gpt 5.6?? fable high is better than XHIGH :(
@reach_vb @xeophon Sol pewnie jak fable wyleci z subskrybcji.
@xeophon Honestly I find fable too expensive for the quality of code it generates anyway. I much prefer to use it as an architect, and find that I get less refusals this way. That said, thank you president trump for even more refusals with this release.
@xeophon Have been thinking abt trying glm5.2 lately
Have liked minimaxm2 and 5.1 was good at cua but not sure how new 5.2 is
Good daily driver?
@reach_vb @xeophon 5.5 xhigh is slow as molasses though
@reach_vb @xeophon codex -m gpt 5.6-sol will be better right?
@reach_vb @xeophon lol, this method really works, bro