Adversarial Audit of RisCanvi AI System

This document is a report detailing the findings of the audit titled 'Automating injustice: an adversarial audit of RisCanvi,' conducted by Eticas. The audit reveals significant deficiencies in the reliability of the RisCanvi tool, which is used in the Catalan prison system to assess inmates' risk of recidivism and violence. The audit methodology included an Ethnographic Audit and a Comparative Output Audit, which highlighted biases in risk classifications based on static factors. Despite regulations in Spain mandating audits for automated systems since 2016, RisCanvi had not been audited until this examination. The report emphasizes the lack of transparency and fairness in the system, noting that inmates often lack legal support and understanding of their risk classifications. Furthermore, professionals using RisCanvi have limited influence on its outcomes, leading to a predominantly automated operation. The audit calls for increased transparency and compliance with the EU AI Act, advocating for fair and unbiased AI technology in criminal justice.