Source record

Published by arXiv cs.HC (T1 source). The original is at https://arxiv.org/abs/2609.13165.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

Single-user longitudinal study (~4,000 conversations over 20 months) introducing 'User-Side Contextual Phenomena' (USCP): three modes — contextual projection, attachment, and authority transfer — showing user-side risk can emerge from long-term AI interaction even when individual responses are acceptable. Offers a non-clinical vocabulary and four evidence roles.

Frames a gap in AI safety evaluation: per-response checks miss risk that accumulates across long interaction. Useful vocabulary for anyone designing or auditing agents meant for repeated use.

Cited by

No citations on record.