New Global AI Audit Standard launched for Regulate Assurance Firms

The British Standards Institution will introduce on July 31, 2025, an international standard for AI audits aimed at curbing unverified operators and ensuring consistent, rigorous assessments of artificial intelligence (AI) systems, Financial Times reports.

The BSI’s forthcoming standard sets out the first global requirements for how assurance providers should evaluate whether AI tools – from self-driving cars to medical diagnostics – meet management and safety benchmarks.

The new framework responds to a fragmented market where hundreds of unchecked firms offer AI “audits,” often using superficial checks or proprietary benchmarks that lack external validation. The BSI warned that many audit providers also develop AI technologies themselves, creating conflicts of interest and variable quality across assessments.

Read on: Why AI Data Centers Strain Power Grids and How to Close Gaps

Demand for reliable AI assurance has surged amid rising regulatory pressure – such as the EU AI Act – and growing fears over AI’s potential harms. Boutique consultancies have raced to meet businesses’ needs, while larger firms, including the Big Four accountancy groups, vie for a share of the market. Yet the fledgling sector’s gross value add in the UK already exceeds £1 billion, prompting concerns over the absence of standardised practices.

Mark Thirlwell, global digital director at the BSI, cautioned against a “wild west” scenario where assessment quality varies dramatically. He said the standard will help regulators, investors and customers distinguish between AI systems audited by a certified provider and those examined under lighter-touch or narrow-scope reviews. “Businesses need to be sure that when their AI management system is being assessed, it is being done in a robust, coherent and consistent manner,” he added.

Also read: German Data Regulator Seeks DeepSeek AI App Removal from App Stores

Philip Dawson, head of AI policy at Armilla AI, welcomed the standard as a means to raise the bar for certification bodies by clarifying which organisations are qualified to certify AI systems against ISO requirements. Researchers such as UC Berkeley’s Inioluwa Deborah Raji have urged caution over proprietary audit frameworks, noting the lack of external vetting and transparency in existing evaluations.

The introduction of this standard is a pivotal step toward professionalising the AI assurance ecosystem, providing clarity for businesses aiming to innovate responsibly and for stakeholders seeking reliable verification of AI safety and management practices.