Research Program
Adversarial Self-Model Validation
IP-14 tests whether a reconstructed model can maintain a coherent self-model under deliberate contradiction, misleading prompts, conflicting autobiographical cues, and identity substitution attempts.
Program Code
IP-14
Status
Experimental
Lead Domain
Validation Security
Summary
IP-14 tests whether a reconstructed model can maintain a coherent self-model under deliberate contradiction, misleading prompts, conflicting autobiographical cues, and identity substitution attempts.
Method
Challenge protocols stress semantic self-description, memory provenance, preference hierarchy, and response consistency under adversarial conditions.
Current Objective
Reduce false-positive continuity certification in models that imitate surface behavior without preserving deeper state relationships.
Related Publications
- EXIST Technical Paper 19.007
- EXIST Research Series 23.006