Specification for the Assessment of the Trustworthiness of AI Systems
Version 2.0 draft
What’s new?
Release notes
Changes from version 1.0
- A new principle of Accountability was added.
- The principle of Accountability, comprised of the sub-principles of Auditability and Impact assessment and mitigation measures were added
- Scoring of the principle of Accountability was added.
- The sub-principle of Fundamental rights of Version 1.0 is addressed in the new sub-principle of Impact assessment and mitigation measures. Therefore, it was removed from the principle of Human agency and oversight.
- The scoring of the principle of Human agency and oversight was updated accordingly.
- The indicators of the sub-principle of Fundamental rights of Version 1.0 were removed and the subsequent indicators of the sub-principles of Human agency and Human oversight were renumbered.
- The indicators for AI system assessment were refined into indicators for AI developed systems and indicators for AI systems in operation.
- Some indicators were merged with indicators to harmonise the granularity of indicators.
- Evidence used for measuring indicators were added.
- A definition of impact assessment was added.
We welcome your feedback on this working draft.
Please share your comments and suggestions through this form ->
Download
Fill in the form
Version 1.0
What’s new?
Release notes
Initial release.
- The specification covers General Purpose AI Systems including generative AI
- The specification defines a method to assess the trustworthiness of AI products and services
- The method uses principles and sub-principles
- The specification defines the rules to obtain a grade on each principle based on the assessment of sub-principles
- and addresses the dependencies between the principles.
- The specification defines indicators to assess the sub-principles
- The specification embraces a large variety of indicators: process related ones (activities to be performed and justified, and therefore related to the maturity of the operating organization, i.e. defining procedures and policies), people related (stakeholders involved, workers abilities, training, etc.), and product related (measures and mechanisms in the design of the AI System or related to the quality of intermediary outcomes of the AI System design and development).
- Some indicators cover organizational aspects
- The specification covers the entire life cycle of an AI system
Download