Claude chats increasingly flag its own guesses with "That is unverified"
ColleenMBrady · x · 2026-10-12
Users are noticing Claude increasingly appending "That is unverified" when offering technical inferences—e.g., explaining a hardware error while explicitly separating speculation from confirmed facts. The behavior has drawn attention as an interesting shift in how the model signals uncertainty.
More from Models
- Users Report Claude Opus 5.5 Silently Routing to a Much Stronger New Checkpoint — rickasaurus · 2026-10-12
- Cognition slide estimates GPT-6 Astra at ~14T parameters, 5x the largest disclosed model — himanshustwts · 2026-10-12
- Can open-source models ever catch up? Reddit debate on one-way innovation flow — CringeyAppple · 2026-10-12
- Claude Max 20x plan now priced at $300/month with $200 in API credits — 0xkarasy · 2026-10-12
- Polymarket puts 62% odds on next Claude 'Fable' model landing within two weeks — Polymarket · 2026-10-12
- ChatGPT randomly slips Hebrew into an answer about a Bleach character — CookiePP · 2026-10-12