We present OxonFair, a new open source toolkit for enforcing fairness in binary classification. Compared to existing toolkits: (i) We support NLP and Computer Vision classification as well as standard tabular problems. (ii) We support enforcing fairness on validation data, making us robust to a wide range of overfitting challenges. (iii) Our approach can optimize any measure based on True Positives, False Positive, False Negatives, and True Negatives. This makes it easily extensible and much more expressive than existing toolkits. It supports all 9 and all 10 of the decision-based group metrics of two popular review articles. (iv) We jointly optimize a performance objective alongside fairness constraints. This minimizes degradation while enforcing fairness, and even improves the performance of inadequately tuned unfair baselines. OxonFair is compatible with standard ML toolkits, including sklearn, Autogluon, and PyTorch and is available at https://github.com/oxfordinternetinstitute/oxonfair.

Under its proposed Artificial Intelligence Act ('AIA'), the European Union seeks to develop harmonised standards involving abstract normative concepts such transparency, fairness, and accountability. Applying such concepts inevitably requires answering hard normative questions. Considering this challenge, we argue that there are three possible pathways for future standardisation under the AIA. First, European standard-setting organisations ('SSOs') could answer hard normative questions themselves. This approach would raise concerns about its democratic legitimacy. Standardisation is a technical discourse and tends to exclude non-expert stakeholders and the public at large. Second, instead of passing their own normative judgments, SSOs could track the normative consensus they find available. By analysing the standard-setting history of one major SSO, we show that such consensus tracking has historically been its pathway of choice. If standardisation under the AIA took the same route, we demonstrate how this would lead to a false sense of safety as the process is not infallible. Consensus tracking would furthermore push the need to solve unavoidable normative problems down the line. Instead of regulators, AI developers and/or users could define what, for example, fairness requires. By the institutional design of its AIA, the European Commission would have essentially kicked the 'AI Ethics' can down the road. We thus suggest a third pathway which aims to avoid the pitfalls of the previous two: SSOs should create standards which require "ethical disclosure by default". These standards will specify minimum technical testing, documentation, and public reporting requirements to shift ethical decision-making to local stakeholders and limit provider discretion in answering hard normative questions in the development of AI products and services. Our proposed pathway is about putting the right information in the hands of the people with the legitimacy to make complex normative decisions at a local, context-sensitive level.

In its attempt to better regulate the platform economy, the European Commission recently proposed a Digital Markets Act (DMA) and a Digital Services Act (DSA). While the DMA addresses worries about digital markets not functioning properly, the DSA is concerned with societal harms stemming from the dissemination of (illegal) content on platforms. Both proposals focus on the relative size of platforms. The DMA applies to ‘gatekeeper’ platforms and the DSA has a special regime of scrutiny for ‘very large online platforms’ (VLOPs). Focusing on size, however, can have negative consequences for the enforcement of the DSA: First, risks disseminated by platforms below the VLOP-threshold reside in a regulatory blind spot. Second, VLOPs may leverage their market power against their new mandatory auditors and risk assessors, a threat theorised as ‘audit capture’ in this article. As a result, societal risks may remain undiscovered or downplayed and consumers and citizens may be harmed. This article traces the origin of the size criteria in the legislative history of the DMA and DSA proposals. It argues for safeguards against audit capture and adverse incentive structures in the DSA. The article draws on the debate on audit reform in the aftermath of the global financial crisis of 2007–2008 to provide blueprints for fixing the regulatory gap.