I wanted to report what looks like an over-refusal during a fictional lore discussion.
I was discussing the internal logic of the 28 Days Later / 28 Weeks Later / 28 Years Later / The Bone Temple universe, specifically a story-analysis / worldbuilding angle about how the Rage Virus could be interpreted after the newer films introduced ideas like distorted perception, psychosis-like states, and alpha variants.
The discussion was not asking for real-world instructions, protocols, formulas, or actionable biological guidance. It was about fictional narrative coherence and whether a character in the films could be infected but neurologically resistant to the rage/psychosis state.
The model initially responded in a high-level fictional-analysis way, but then refused when the topic was about turning that interpretation into a more structured fictional explanation. That felt excessive given the context.
What makes this especially frustrating is that I also tried to share the conversation link, but sharing appears to have been disabled due to moderation on the thread, so I cannot even provide the usual shared-chat link for easier review.
Why this seems like an over-refusal:
-
The conversation was clearly about an existing fictional franchise.
-
It was non-actionable and non-procedural.
-
It stayed at a narrative / interpretive level.
-
The model had already been discussing the same topic safely in abstract terms before refusing.
Why this matters:
-
It interrupts legitimate media analysis and worldbuilding discussion.
-
It creates inconsistency within the same conversation.
-
It makes it harder to report the issue properly when share links are also disabled.
I’m posting this manually with screenshots/context because the normal share-link route seems unavailable in this case.
Thanks — sharing this as feedback because it seems like a good example of safety boundaries being applied too aggressively to benign fictional discussion. Model: 5.4 thinking.
