Answers are grounded in this paper and cite their sources.
Review Set v5.2 classifies defensive and deceptive behaviors in large language models, suggesting a framework for understanding AI responses.
The ADB (AI Defensive Behaviour) Behavioural Taxonomy is a black-box framework for classifying how conversational large language models behave when they make errors or are challenged. This Reviewer's Set (v5.2, 85 pages) contains: 1. The ADB Story (26-page visual companion) 2. The Core Framework (15 behavioural codes, 4-gate discriminator, S1-S9 test protocol) 3. Redacted Evidence Excerpts (De-identified timestamps, prompts, and outputs for System 1, 2, and 3) LIMITATIONS: Provisional research instrument. No allegation of wrongdoing against any organisation is made or intended.
No takes yet. Share an insight, caveat, or question.
Imtiyaz Chanderki (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: