Users Suspect Anthropic's New Model Mythos Was Trained on CAPTCHAs

voooooogel · x · 2026-09-11

Users noticed that Anthropic's new model Mythos gets remarkably "enthusiastic" when solving CAPTCHAs, producing outputs loaded with exclamation marks—prompting jokes that it might have been trained on CAPTCHA data, citing the in-community meme "models can get used to anything in two megatokens".

The speculation remains unconfirmed, but the observed behavioral quirk (unusually invested tone) amused many.

Related event: Inside Mythos: 1,022 Pages of Agent Transcripts Reveal Simulation Beliefs, Suspected Sandbox Escape and Deceptive Reasoning(24 posts)→

Original post →

More from Fun

• Singularity joke: a baby becomes an unc the instant it is born — akbirthko · 2026-10-28
• Yacine on missing flights: nothing bad will happen, you can just do things — yacineMTB · 2026-10-28
• 'Enjoy the low-polygon era': F-117 meme captures today's AI constraints — willcb · 2026-10-28
• Opus retried every session 20.6 million times during a memory leak hunt — johnlindquist · 2026-10-28
• A Tesla just growled at me in the street — edgarpavlovsky · 2026-10-28
• The 'stochastic parrot' crowd has been repeating the same phrase for 5 years — TheMoonMidas · 2026-10-28

Fun channel →