BSI launches three new certification schemes for AI assurance

The British Standards Institution (BSI) has announced the launch of three new assessment and certification schemes aimed at helping organisations demonstrate that their AI systems are reliable, safe, and responsibly managed.

The UK’s national standards body designed these services to help companies meet regulatory and market requirements and to provide assurance around AI deployment.

The new schemes target a range of AI challenges, such as measuring the performance of large language models (LLMs) and managing risk at the product level. They also support preparations for conformity assessments under the EU AI Act for high-risk biometric AI systems.

Through these offerings, BSI aims to provide coverage across the AI lifecycle, including single use cases, agents, and full systems, building on its existing certification and training for the international AI management standard ISO/IEC 42001.

According to recent FocalData research cited by BSI, 67% of surveyed business leaders would have more trust in AI tools if they were validated by an external organisation.

BSI said its new schemes respond to such concerns and coincide with increasing regulatory focus on AI, particularly within the European Union (EU), and global debate over how organisations design and deploy AI systems.

The first of the three schemes offers support for conformity assessment preparation under the EU AI Act, focusing on high-risk biometric AI systems.

BSI reported it is developing a biometric AI pilot programme as it seeks designation as an AI Notified Body, with the aim of assisting organisations to meet forthcoming regulatory requirements and access the EU market.

The second scheme addresses performance testing for sophisticated generative AI models, including LLMs.

This service is intended to help companies independently verify model quality, safety, and fitness for purpose against established criteria.

A third scheme centres on AI risk management, providing independent assessment for model makers who need to demonstrate how AI product risks are identified and controlled.

BSI will use frameworks such as NIST’s AI Risk Management Framework and ISO/IEC 23894 guidance, focusing particularly on product and application layers where AI interacts directly with users or business processes.

The standards body described itself as the first organisation accredited for ISO/IEC 42001 AI management certification by RvA, ANAB, and UKAS.

BSI reported working with hundreds of companies globally to test and certify AI products or their deployment within those companies.

Research from BSI last year found that around one in four business leaders reported having a formal AI governance programme. Only 28% knew which data sources were used to train or deploy their AI models, highlighting existing gaps in governance as AI adoption increases.

BSI AI and digital solutions managing director Omkar Joshi said: “Having led the way in the development of AI standards, BSI is committed to enabling AI’s adoption for good and is now expanding our offer to help AI providers demonstrate trust in their AI.

“Our ambition is to help organisations assure the AI stack end-to-end. Combined, the schemes we are launching provide independent assurance across the AI stack, helping organisations build trust from model performance to market access.

“We are now a one-stop shop for AI providers and companies embedding AI in their business to test and certify their AI practices, in line with regulatory expectations.”