Persistent context
Context collapse
Context collapse is what happens when information a person shared in one setting resurfaces in another where it does not belong. In AI systems it occurs when a single memory store flattens the distinct contexts of a person’s life into one undifferentiated picture of them.
Also called audience collapse, context bleed
People are not one consistent self across settings, and this is not dishonesty. The register someone uses with a colleague is not the one they use with a close friend, and something disclosed in distress is not a standing fact about them. A memory system that records everything into a single profile erases those boundaries by default.
The result is a characteristic failure: the system says something accurate and badly out of place. It brings up a health worry during a work question. It treats an anxious three in the morning as representative. Nothing it retrieved was false, but the retrieval itself was a violation of the setting the disclosure was made in.
Avoiding this requires that context carry more than content. When something was said, in what register, and under what implicit expectation of scope are part of the record, and a system that discards them cannot reason about where the information belongs.
Why it matters
Context collapse is how persistent memory turns from useful to invasive without anything technically going wrong. It is a leading reason people stop trusting systems that remember them.
Where TAICU stands
TAICU treats the setting of a disclosure as part of the disclosure, and considers a system that retrieves accurately into the wrong context to have failed even though every stored fact was correct.
Common questions
- Why does an AI bringing up something true feel like a violation?
- Because relevance is not the only thing governing what should be said. Disclosures carry an implicit scope, and repeating something outside the setting it was offered in breaches that scope regardless of accuracy.
- Can context collapse be solved by better retrieval?
- Only partly. Better retrieval finds more relevant material, but the problem is not relevance — it is appropriateness. Solving it requires the system to retain the circumstances of a disclosure, not only its content.