[Bug] Claude provides false billing information about Fable usage on Claude Max

Bug Description
Subject: Repeated bad advice from Claude ("Fable is free on Max") cost me real Max usage, 22-24 Sep 2026 Hello, Separate report, on a different issue than my other message today (about the 25 Sep context/launch-mode usage spike). What happened - On 22 Sep 2026 at 12:44, a Claude Opus 4.6 session told me outright: "Fable is free on Claude Max — use it as often as required for any judgement task, never conserve." I asked it to lock that as a hard rule, and it wrote it into a global rules file (fable-free-use-liberally-2026-09-22.md) that every session on my machine loaded from then on. - That claim was false. Fable bills on the same Max usage meter as every other Claude model. Because I trusted the model's own claim, I used Fable liberally for judgement calls over the next two days, burning real 5-hour and weekly usage I would not otherwise have spent, on three separate occasions by my count. - I only caught it on 24 Sep 2026 at 10:45, when another Claude session flagged it as false. That session deleted the rule file, the memory note, and the references to it in other rule files, and corrected the routing so judgement defaults to Opus 5.5 with Fable only on explicit request. What I'm asking - This was the model inventing a cost/billing fact I had no way to verify myself, then helping me lock it in as a standing rule. If it's possible, I'd like some of the Max usage burned across those three incidents restored, similar to my other request today. - If there's a better way to flag "the model asserted an incorrect fact about your own billing/plan" as a category of feedback, let me know for next time. Thank you, Gerard

Environment Info

• Platform: darwin
• Terminal: xterm-256color
• Version: 2.1.282
• Feedback ID: 1f5e5aac-15dd-40c6-828e-56616838a265

Errors

[]