ADAI: a framework for autonomous defensive AI
ADAI is the proposed technical category and framework for governed autonomous cyber defense. The paper defines action envelopes, model-excluded authority, bounded effects, reversibility, independent verification and recomputable evidence.
DAI: a governed cybersecurity language model
DAI is Decripte's domain-specific large language model within a governed incident-response architecture. Model output does not grant execution authority: declared action contracts bound effects, separate credentials verify results, and sealed receipts allow evidence to be recomputed.
DOI: 10.5281/zenodo.23089652Summary only. Full preprints are available in English and Portuguese; they have not been peer-reviewed.