What This Looks Like
The AI uses older context that should no longer govern the task. The output may reflect a prior instruction, old file version, previous decision, outdated source, earlier user preference, or abandoned task state even though the user expects the current request to be based on newer context.
Why It Matters
Stale context can be hard to detect because the output may still look coherent. The user has to notice that the answer is governed by an older state rather than the current one. This can cause wrong recommendations, outdated references, incorrect formatting, or workflow decisions based on context that should have expired.
Structural Signal
A prior context state continues to influence the output after the governing context has changed. The issue is not that context is missing; it is that the wrong context remains active beyond its valid boundary.
Common Triggers
- A prior instruction is never explicitly revoked
- A file, source, or policy changed but the old version remains in context
- The AI treats earlier task state as still active after the user changes direction
- Saved memory or conversation state is broader than the user expects
- A restore or reload preserves outdated context
- The workflow lacks a rule for replacing old context with newer context
When to Use This Issue
Use this Issue when old context continues to shape the AI output after it should have been replaced, expired, narrowed, or ignored.
When Not to Use This Issue
Do not use this Issue when the AI lacks needed context entirely, or when the user intentionally asks the AI to reuse prior context. This Issue applies when stale context is active and materially affects the output.