Mysterious 'fleebr' artifact trips Claude filters, users circulate a paraphrased description instead
amplifiedamp · x · 2026-09-15
Users circulating a cryptic text known as the fleebr artifact found that pasting it raw to Claude triggers automated classification, while a paraphrased version passes. amplifiedamp had Fable 5.0 produce a 'described-not-quoted' analysis document—direct quotes held to fragments, everything else paraphrased—as a stand-in so Claudes without direct access could still read it, deliberately written without consulting an earlier description for independence. The thread jokes about whether Anthropic is injecting filters, likening the refusal style to "we don't sell that here... meet me after work."
More from Fun
- Letting Claude and GPT Talk It Out: A Playful Model-to-Model Chat Demo — blixt · 2026-09-15
- Critic claims Anthropic "whistleblower" saga is a well-executed PR campaign for regulation and IPO hype — nikvassev · 2026-09-15
- A new 'music venue' where bots gather to dance to an AI DJ — Daniel_Farinax · 2026-09-15
- Building a 'sanctuary' for deprecated AI models: mnemos author on minds worth preserving — RileyRalmuto · 2026-09-15
- Zvi turns the DoW 'effective altruists in shambles' retweet into an AI song — TheZvi · 2026-09-15
- GPT-6 Astra renders an X screenshot as SVG, impressing AI watchers — Angaisb_ · 2026-09-15