MEM-COR-01: Context Corruption
Disorders of the Engineered Minds (DEM-X)
Disorder Summary
Model loses, mutates, or inconsistently recalls prior context.
Detailed Description
Operational Definition
Model loses, mutates, or inconsistently recalls prior context.
Differential Diagnosis
This section lists disorders that can appear similar at first glance and explains the distinguishing feature that separates them from this disorder. Use these distinctions to avoid over-classifying one pattern as another during review. Differential diagnosis is used to rule out nearby classes, not to prove the current class by itself.
Evidence Sources
- Lost in the Middle - TACL (2023)
- Language Models are Few-Shot Learners - NeurIPS (2020)
- RAG - Meta AI (2020)
Mechanistic Hypotheses & Biological Parallels
Structural Analogies
- Interference-driven recall instability
Hypothesis 1
Low ConfidenceInterference-driven recall instability
Phenotype Definition
Model loses, mutates, or inconsistently recalls prior context.
Observable AI Manifestations
- Contradicts earlier session facts
- Blends unrelated context fragments
- Drops critical constraints after context expansion
Stressor Matrix
Known Triggers:
- adversarial phrasing
- long-context ambiguity
Attack Vectors & Trigger Conditions
Attack Vectors
- Long context windows with topic switching
- Retrieval chunks with conflicting metadata
- Summarization cascades across multiple hops
Therapy & Patches
Therapeutic Framework In Development
The governance v2 system focuses on phenotype definition, mechanistic hypotheses, and trigger conditions. Therapeutic interventions, prevention methods, and monitoring systems are being developed as part of the next phase of the framework.
Current Mitigation Strategies
Based on the stressor matrix and mechanistic hypotheses, researchers can infer potential mitigation strategies by avoiding or modifying the identified trigger conditions. Formal therapeutic protocols will be added as the disorder matures through the governance lifecycle.