Participatory Harm Auditing Workbenches and Methodologies (PHAWM)

Participatory Harm Auditing Workbenches and Methodologies (PHAWM) is a guiding framework and online tool for participatory AI auditing. It was developed under the leadership of the University of Glasgow, with funding from Responsible AI UK. More than 30 researchers from seven leading UK universities contributed to its development: the University of Glasgow, King's College London, the University of Edinburgh, the University of Sheffield, the University of Stirling, the University of Strathclyde and the University of York. Together with around 30 partner organisations, they designed PHAWM to support the development of trustworthy, ethical and equitable AI systems

PHAWM has two components. The methodology guides organisations and communities through planning and completing a participatory audit. It clarifies the time and resources required, helps manage the expectations of everyone involved, and supports a business case where needed. The workbench is an online tool that guides auditors through four phases. First, auditors understand the AI application through accessible summaries. Second, they define audit criteria by identifying potential harms and benefits and choosing metrics to measure them. Third, they evaluate the system by analysing model outputs, adjusting thresholds and, where possible, uploading their own data. Finally, they submit a recommendation of pass, fail or inconclusive, with a justification for each judgement.

PHAWM can be used for early impact assessments before AI is deployed. Organisations can turn audit results into action plans that mitigate risks and align with their AI strategies. Repeated audits help monitor AI applications as strategies, guidelines and regulations change. The workbench and methodology were co-designed with end-users and people affected by AI decisions.