I'm just wondering how Mythos and Fable are defined, exactly, for the internal employee restrictions?
Vision Transformer researcher Lucas Beyer sparks a vague online exchange over internal rules referencing "Mythos" and "Fable
Story Overview
Meta researcher Lucas Beyer, known for Vision Transformer work, posted a direct question about how the capitalized terms Mythos and Fable are defined under internal employee restrictions, only to receive a curt “Ssshhh” reply from academic Dan Roy.
Precise scope of the restrictions stays hidden
No public details clarify whether Mythos and Fable refer to model variants, usage policies, or other internal boundaries, leaving the exact reach of the rules unstated.
Colleague reply hints at non-public sensitivity
Roy’s single-word response implies the terminology touches on matters not meant for open discussion, though the thread supplies no further context or confirmation.
Many users criticized Anthropic's Mythos and Fable internal restrictions as deceptive, hypocritical toward regulators, and overly controlling.
Most Activity

@giffmana i think they can't access it if they're not US citizens... probably they've shut down access to all until clarification
@giffmana Ssshhh.
I'm just wondering how Mythos and Fable are defined, exactly, for the internal employee restrictions?

@XMihura My question is, what exactly, unambiguously, is "it"

@giffmana The worst thing is that everybody, including their lawyers will have the same question (and no answer) and you can guess how that will go with the said lawyers.

@giffmana Not a problem, can just use "mythos-v4ckpt-SFT-final2-gov-safe-aligned"

@giffmana aha, unambiguously
I guess at least all checkpoints derivated from whatever weights they used to make the system cards (?)

@giffmana doesn’t the order explicitly name “Mythos” and “Fable” — they’ll just have to use the unreleased models still using codenames of course

@giffmana For a CEO who already went for a pissing contest with the USG twice I would find it very hard to justify annoying the USG yet again and seeking the next head on confrontation. Dismantling and nationalization are faster on the table than Dario can spell apology.

@giffmana They make the internal employees pink promise not to use it.

@pastaraspberry Yeah

@giffmana I wonder if they eventually just end up banning models with more than X trillion parameters

@giffmana at Anthropic it’s called “infernal employee restrictions”

@giffmana 他们根本不会限制,哈哈; 他们对外满嘴仁义道德和法治,对内拒绝遵守法律监管

@giffmana that sounds like an interesting rabbit hole, how do they differentiate?

@giffmana the sim-to-real gap. that's the one problem that never gets solved, just renamed.

@giffmana us citizen ok, maybe. for foreign employees mmm based on clearance, historically speaking. everytime the government tried to absorb a new tech, that’s how they did it by controlling who had access to it:))) so yup, labs lost control over their own models:( pretty sad actually

@giffmana @XMihura mythos_epoch99_final_reviewed.pt

@giffmana it’s even more narrow — Mythos 5 and Fable 5

@XMihura @giffmana It’s quite possible the line between this and Opus is not clear

@giffmana Your answer is in ITAR/EAR policy (not legal advice).
It's a good read, if you're into that sort of stuff.