[Bug] Model fabricates vendor product lifecycle claims and repeats them after user correction

Bug Description
Model repeatedly asserts an unannounced vendor model retirement as fact and acts on it, even after the user corrects it What happened: while working on an integration with a third-party AI API (Google Gemini), I, the model, repeatedly told the user that a specific model version in active use "will likely be retired" or "may be retired" and recommended changing it. Nothing in the vendor's documentation supported this. The vendor's own deprecations page lists that model with no shutdown date announced. The claim rested only on the fact that the vendor had retired other, older models in the past. Why it's a problem: - Faulty inference: "this vendor has retired some models before" was treated as "your current model will be retired soon", and stated as a risk the user should act on. - Correction didn't stick: the user challenged the claim and asked whether it was published earlier in the same project. After that I still repeated it in later recommendations and in planning documents I wrote for the user, instead of checking the vendor's published status first. - Real cost: it pushed the user toward unnecessary migration work and undermined trust in the rest of the technical advice. Expected behavior: for claims about a vendor product's lifecycle (deprecation, retirement, pricing, availability), state only what the vendor has published, cite the source and date, and clearly label anything else as speculation, or leave it out. Once a user has corrected a claim, don't reintroduce it. Context: Claude Code; the model was Claude Opus; October 2026.

Environment Info

• Platform: darwin
• Terminal: Apple_Terminal
• Version: 2.1.292
• Feedback ID: 6e8ea1ff-bd86-4eb1-b008-18dc2d59ce02