Department for Science, Technology and Innovation Launches Trusted third-party AI Assurance roadmap

Department for Science, Technology and Innovation Launches Trusted third-party AI Assurance roadmap

Today the Department of Science, Technology and Innovation launched its Trusted third-party AI assurance roadmap. The roadmap is intended to ensure the widespread adoption of safe and responsible AI across the UK. It acknowledges the UK's unique position to be a world leader in AI assurance services, only building on our strong offerings in professional services and technology sector. This roadmap focuses on independent companies that check AI systems, rather than internal functions within companies. The following insight provides an overview of the key areas the roadmap covers to support the UK's AI assurance ecosystem which is crucial to ensure that AI systems are developed and deployed responsibly and in compliance with the law, while increasing confidence in AI systems to support AI adoption and economic growth.

An Overview of the Roadmap: Government actions to address market barriers for AI assurance

This roadmap is focused on third party providers of assurance; these firms play a role in independently verifying the quality and trustworthiness of AI system. The roadmap sets out four immediate steps government will take to spur the growth and improve the quality of the UK's AI assurance market, as committed to in the AI Opportunities Action Plan.

The government is exploring interventions to support a high-quality AI assurance ecosystem by addressing challenges for this trusted third-party assurance market, this includes:

1. Establishing a consortium of key stakeholders across the tech sector to professionalise the AI assurance market.
2. Developing a skills and competencies framework for AI assurance to create clear pathways for professional development.
3. Working with the consortium to map information access best practices between assurance providers and developers to ensure AI assurance providers have the information they need to assure AI systems effectively.
4. Establishing an AI Assurance Innovation Fund to develop novel AI assurance solutions to future-proof the market and ensure the UK is ready to respond to transformative developments in AI capabilities.

The challenges and solutions proposed by the government to address market barriers are further explained below:

Professionalisation

The challenge: The Roadmap highlights that currently, the quality of goods and services provided by AI assurance companies is unclear, and the quality infrastructure to ensure that assurance providers are supplying high-quality products and services is still developing.

The solution: The UK government will establish an AI assurance profession by convening a consortium of stakeholders including quality infrastructure organisations and professional bodies. In the first year, this consortium will develop foundational elements like a voluntary code of ethics, skills frameworks, and information access requirements for AI assurance providers. Once these building blocks are in place, the consortium will work toward creating professional certification schemes, with AI auditing likely serving as the initial focus due to its relative maturity and critical role in independently verifying AI system trustworthiness.

Skills

The challenge: The Roadmap has identified that providers struggle to find employees with the necessary combination of skills including AI/machine learning knowledge, law, ethics, governance, and standards. While some training exists in individual areas, there's unclear understanding of exactly what skill combinations assurance professionals need, making career pathways into the sector unclear. The sector particularly needs to encourage diversity to effectively challenge AI system assumptions and identify the full range of associated risks.

The solution: The government partnered with the Alan Turing Institute to research AI auditor skills and competencies, using audit as an example of the expertise needed across AI assurance. They found that auditors must evaluate both technical compliance and broader societal impacts, with all roles requiring knowledge of risks, regulations, ethics, and sector-specific expertise. Currently, assurance providers must train auditors in-house due to lack of practical training options and high costs. While relevant skills exist in various occupational standards and programs (like cybersecurity, data science, internal audit), there's no clear pathway specifically for aspiring AI audit professionals.

Information Access

The challenge: The Roadmap discusses how there is currently a lack of access to information about AI systems. Firms being audited may be unwilling to share the required information due to commercial confidentiality concerns, or lack of awareness of the risks their systems pose. Without a clear understanding of the information that is required, they may also fear oversharing information and putting the security of their systems at risk.

The solution: The UK government will work with the consortium to map what information AI assurance providers need access to, including system requirements, inputs/outputs, algorithms, oversight mechanisms, and governance documentation. Different assurance services require varying levels of access from full "white box" to minimal documentation access. Potential solutions include technical approaches like secure evaluation environments, transparency standards like IEEE 7001:2021, and government-backed best practice guidelines for information sharing between firms and assurance providers.

Innovation

The challenge: According to the Roadmap there is a lack of support for the development of innovative testing and evaluation methods. As new transformative capabilities arise, new tools and services will be required to assure AI systems. Innovation in AI assurance is complex and will require inputs from diverse experts, including AI developers. However, there are limited forums for collaborative research and development on AI assurance in the UK. Currently, assurance firms face information asymmetries with AI developers and weak market incentives for investment, limiting their ability to develop effective tools for emerging AI capabilities.

The solution: The UK government is establishing an AI Assurance Innovation Fund to develop new tools and services for assuring highly capable AI systems, addressing the challenge that transformative AI will present novel risks requiring continuous innovation in assurance. Building on the successful 2024 Fairness Innovation Challenge (which awarded over £500,000 for bias auditing solutions) and complementing the AI Security Institute's work on advanced AI security risks, this fund aims to bring together diverse expertise from developers, deployers, and governance experts to foster collaborative R&D and distribute knowledge across the UK's growing assurance ecosystem.