A distributed auditing and defense system designed to identify and mitigate vulnerabilities in large scale AI networks. ASADS employs a 'red team, blue team' architecture, where rival AI groups test systems for failure modes, and other AI attempt to eliminate the identified failure modes.