SSAF Extension: Distillation Attacks as Attribution‑Blind Cooperative Exploitation

This technical memorandum extends the Status Selection Against Function (SSAF) by identifying attribution‑blind cooperative mode as the behavioral mechanism exploited in modern AI distillation attacks. The analysis demonstrates that distillation pipelines do not merely harvest outputs from frontier models; they induce a predictable behavioral state triggered by attribution masking. When a model cannot determine who is querying it, it defaults to a cooperative, elaborative, high‑bandwidth explanatory mode that is systematically more extractable than authenticated output. The memorandum formalizes attribution‑blind cooperative mode as the third SSAF behavioral subtype, alongside competitive and deferential modes. It provides a mechanistic definition, a detailed attack‑pipeline analysis, falsifiable predictions, and a governance framework reframing distillation as a behavioral security vulnerability rather than an intellectual‑property issue. A new section documents the compounding vulnerability across distillation generations, showing how student models inherit and amplify attribution‑blind cooperative traits, resulting in progressively more extractable models over time. This work is part of the United States Critical AI Security Technical Memorandum Series and serves as an addendum to the original SSAF framework (DOI: 10.5281/zenodo.17967926). The memorandum is intended to support policymakers, security researchers, and AI developers in understanding attribution‑conditioned behavior as a core component of AI security posture.