The BSI has published a new international technical specification designed to make artificial intelligence (AI) systems more transparent and to protect users from unexpected impacts.
The document, Making AI Systems Decision-Making Transparent (ISO/IEC TS 6254), offers organisations practical guidance for creating explainable AI systems, those whose outputs and decisions can be understood by humans.
The new specification provides a clear framework to help ensure that AI used in decision-making can be scrutinised and trusted by users across all sectors. As AI increasingly informs critical decisions in areas such as healthcare, finance and criminal justice, the guidance aims to build public confidence and support ethical, responsible innovation.
The BSI’s research has shown that more than half of people believe vulnerable users need greater protection to ensure they can benefit from AI technologies, and nearly two-thirds want a standardised way to flag concerns, errors or biases. According to the BSI, embedding transparency throughout the AI system lifecycle is essential to achieving these goals.
“As AI continues to shape high-stakes decisions across our society, transparency is no longer optional,” said David Cuckow, BSI director, “it’s essential and critical for AI being a force for good in society. This new specification provides a vital framework to ensure AI decisions can be understood, scrutinised and trusted. It marks an important step towards building ethical, responsible AI systems that serve people and society fairly.”
Developed through international collaboration and informed by real-world use cases, ISO/IEC TS 6254 builds upon the progress made by tools such as Google’s What-If Tool and IBM’s AI Fairness 360 toolkit, which already help improve the interpretability and fairness of AI models.
The new guidance provides a consistent, globally recognised approach to designing and assessing transparent AI systems and complements policy initiatives including the UK Government’s Artificial Intelligence Playbook, which calls for AI to be “as explainable as possible.”
The specification complements BS ISO/IEC 42001 – Information technology. Artificial intelligence. Management system, published in 2023, which supports responsible AI use and has already been adopted by several major organisations worldwide.
For more information or to purchase the standard, visit the BSI Knowledge platform.
