Authors: Rokas Gipiškis, Ayrton San Joaquin, Ze Shen Chin, Adrian Regenfuß, Ariel Gil, and Koen Holtman
This paper provides the most extensive documentation to date of risks and safety measures for general-purpose AI (GPAI) systems, offering critical support for global AI regulation and safety standards efforts. The catalogue systematically documents risk sources, including technical, operational, and societal risks across the entire AI lifecycle, alongside established and experimental methods for identifying and mitigating these risks.
The research spans critical domains including model development challenges, evaluation limitations, AI safety concerns like adversarial attacks and deception, deployment risks, and broader impacts on cybersecurity, society, privacy, and the environment. This comprehensive approach helps AI providers, regulators, researchers, and policymakers identify systemic risks that might otherwise be overlooked and implement state-of-the-art risk management practices.
The catalogue has been released under a public domain license, allowing anyone to use, adapt, or incorporate the content directly into regulations, standards, or safety frameworks with or without attribution.