BSI Algorithmic Auditing

BSI is a businesfoss improvement and standards company. With over 100 years’ experience in shaping the development of consensus-led best practice based guidance and delivering independent conformity assessments, BSI is committed to exploring agile ways of building consumer trust in innovations that have the potential to make a positive impact on our way of life by incorporating AI and other emerging technologies.

Market driver

The European Union (EU) is addressing the issue of high-risk artificial intelligence (AI) in software and hardware devices through the development of new regulation, the EU AI Act.

BSI is now progressing towards achieving AI-notified body status. This requires a robust audit review process to enable the approval of AI applications in medical devices and other products and services against the new regulations.

To support this journey, BSI was keen to embrace the leading technology at the cutting edge of algorithmic auditing. Therefore, BSI partnered with L Marks to run a multi-step innovation programme to help identify a leading start-up to carry out the auditing of AI models at scale, as part of their audit review process.

The approach

BSI set two challenges for the shortlisted start-ups to address over an initial three-day Fusion event. These challenges involved testing the applicability of the solutions and finding opportunities to accelerate the build.

Four teams were selected to progress based on the potential of their products and openness to adapt to the programme requirements. Two further three-week sprints tested the teams’ solutions against technical requirements and non-functional requirements, to determine which was most suitable.

The four participating teams were:

• Fairly AI – Compliance automation for AI systems, designed to help get AI to audit-ready status fast
• Dragonfly – A new blueprint for managing engineering and quality teams
• Citadel AI – Provides real-world monitoring, testing, and simulation for models and datasets
• KOSA AI – state-of-the-art technology enabling people to understand and mitigate AI bias and reduce risks

The two sprints created an innovative space to problem solve, test, refine, and develop the commercial viability of the AI audit algorithms in a collaborative way. Finalists had the opportunity to:

• demonstrate the versatility and adaptability of their team and technology in addressing BSI’s challenges.
• develop potential roadmaps for the robust deployment of an algorithmic auditing solution.
• demonstrate commercialisation of the solution from resource requirements and knowledge gaps to areas of risks and opportunity.

Results:

Following a Final Review at the end of the programme, Citadel AI was selected as the BSI’s preferred supplier. Citadel’s algorithm auditing delivers robust technical capabilities, strong integrations with technical standards and a user-friendly interface, which gives a commercially scalable solution for the future of AI audits. This process was a great way to identify new automated methods to enhance the audit review process using the latest technology.

Through Citadel AI’s tools, as part of the assessment process, BSI can measure AI compliance against technical standards, supported by in-depth technical analysis including fairness testing, bias detection and robustness testing. The tools can also automatically generate comprehensive analysis and reporting for supporting the assessment process, assuring the safety and reliability of AI systems.

BSI and Citadel AI will initially focus on the healthcare sector and subsequently expand their services to oversee other higher-risk sectors such as the automotive industry and other highly sensitive products such as biometric identification, helping to protect individual identities. BSI is committed to adopting these cutting-edge tools to support us in solving society’s biggest issues – from improving the quality of life for patients with diabetes to protecting individual identities – in a safe and ethical way.