Persistent context
Context collapse
Context collapse occurs when information shared in one setting resurfaces in another where it does not belong. In AI systems, one memory store can flatten the separate contexts of a person’s life into a single, undifferentiated picture.
Also called audience collapse, context bleed
People do not present one fixed self in every setting, and that is not dishonesty. The register used with a colleague differs from one used with a close friend. Something said in distress does not become a standing fact. A memory system that puts everything into one profile erases those boundaries by default.
The result has a familiar shape: the system says something true in a badly misplaced moment. It raises a health worry during a work question or treats an anxious 3 a.m. exchange as representative. The retrieved material may be accurate, yet retrieval breaches the setting in which it was shared.
Avoidance requires context to retain more than content. When something was said, in what register, and with what expected scope all belong in the record. A system that drops those details cannot judge where the information belongs.
Why it matters
Context collapse lets persistent memory become invasive while every stored fact remains technically correct. It is a common reason people stop trusting systems that remember them.
Where TAICU stands
TAICU treats the setting of a disclosure as part of the disclosure. Accurate retrieval in the wrong setting is still a system failure.
Common questions
- Why can an AI repeating something true feel like a violation?
- Accuracy alone does not govern what belongs in a conversation. Disclosures carry an expected scope. Repeating one outside the setting where it was offered breaches that scope.
- Can better retrieval solve context collapse?
- Better retrieval can find more relevant material. Context collapse concerns appropriateness as well. The system must retain the circumstances of a disclosure alongside its content.