IAD, zero-retention biometrics can bridge EU-UK governance gap

Threats tend to wax and wane, especially in tech, where constant innovation drives advances in fraud. Early in 2025, the UK government declared deepfakes to be “the greatest challenge of the online age” – a scourge requiring urgent attention from legislators and biometrics providers. Deepfake detection was about to come into its own.

A year and half later, deepfake detection may be preparing to join the list of technologies that are so integrated into larger models that they become, in themselves, unremarkable. Think, for instance, of touch screens: once a marvel, now so standard-issue that to not have one is unusual.

That’s not to say deepfakes have gone away. Rather, as tech has evolved, so too has our understanding of how deepfakes fit into the larger fraud picture. Cheap, free and easy to spread with the right tools, deepfakes are beginning to look less like a complete weapon and more like ammunition to be launched using other, related fraud techniques – most notably, injection attacks. Yet the regulatory frameworks are not yet in place to handle what this shift implies.

Dr. Felipe Romero-Moreno, associate professor at the Hertfordshire School of Law, Education and Society, argues as much in a new paper. “While the industry continues to heavily rely on Presentation Attack Detection (PAD), my peer-reviewed findings reveal that this legacy control is rapidly failing against the evolving threat landscape,” says Romero-Moreno in a letter to Biometric Update. “Threat actors are now actively bypassing physical camera sensors, driving a 2,665 percent surge in digital injection attacks.”

The article, “The enforced technical mandate: A multi-layered governance model for deepfake fraud and biometric integrity,” proposes a new governance architecture to solve the problem of fragmentation across regulatory regimes. Romero-Moreno argues that the current approach is “trapped between the European Union’s rights-based architecture (General Data Protection Regulation, AI Act, Digital Services Act) and the United Kingdom’s safety-oriented technology-forcing imperatives (Online Safety Act).”

“While existing literature addresses specific dimensions of the deepfake threat (e.g., algorithmic detection, criminal law implications), a pivotal oversight deficit remains: current research lacks a comprehensive, integrated analysis bridging technical fraud prevention and the multi-layered legal lifecycle.”

A critical gap exists between the EU and UK’s approaches, Romero-Moreno says – and the UK’s “distinct technology-forcing directives regarding financial deepfake fraud” create “critical conflicts with privacy law.”

Paradigm identifies three layers of governance

Romero-Moreno’s thesis is that “the legal response to deepfake fraud has crystallized into a unified, three-layered governance model, where legal enforcement and technical defence are inextricably intertwined.” His analysis moves through three governance layers that constitute his paradigm:

Layer 1 (Source Control) examines “the transition of Privacy-Enhancing Technologies (PETs) from discretionary practices to binding legal obligations,” scrutinizing the regulatory frameworks governing AI architecture and biometric inputs – the “raw material of deepfake fraud” – to formulate “a legal counter-narrative regulating the technology at its foundational inception.”

Layer 2 (Distribution Control) contrasts two regulatory modalities: the EU’s systemic risk management under the DSA, and the UK’s “unique technology-forcing requirements under the OSA, which compel proactive fraud detection.”

Layer 3 (Accountability) “evaluates ex-post enforcement, including criminal and corporate liability.”

In short, then: what governs the digital stuff of biometric fraud, what regulates its distribution, and what happens when someone breaks the rules.

Make a (biometric) hash of it

The paper works through these layers to arrive at its conclusions, which include the call for advanced injection attack standards. Key axioms note the “accountability lacuna” that results when laws lag behind change, and the growing threat of “Frankenstein” synthetic identities.

Regarding defensive integrity – Layer 2 – it says “current algorithmic compliance proves insufficient against the 2665 percent surge in digital injection attacks. This mandates a shift to high-assurance Injection Attack Detection (IAD) standards.”

The fundamental binary – EU versus UK – appears again in the conclusion, as an issue in need of resolution. “While the EU emphasizes the rigid processing of biometric data and the UK prioritizes technology-forcing duties, the implementation of a financial blockade compels a unified technical front,” Romero-Moreno says.

As a solution, he proposes amendments to the ISO 20022 standard on financial messaging and data exchange. Specifically, Romeno-Moreno argues that the enforcement of NIST IAL2 zero-retention biometric standards, and the integration of biometric integrity hashes directly into ISO 20022, can create an automated blockade against industrialized Fraud-as-a-Service networks.

“This research contends that integrating biometric integrity hashes into the ISO 20022 messaging schema serves as the pivotal technical bridge between these bifurcated philosophies. This approach reconciles the EU’s verify-then-publish mandate with the UK’s safety duties by embedding legal compliance directly into the foundational plumbing of the global digital economy.”

Further policy recommendations include mandating NIST SP 800-63-4 Identity Assurance Level 2 (IAL2) as the statutory baseline, which means committing to zero-retention biometrics, “wherein raw data deletes immediately following a 1:1 match,” and cryptographic biometric integrity hashes serve the purpose of an identifier.

IAD a core financial risk mitigation imperative

On enforcement, “regulators must institute a transparent, statutory liability and redress system modelled on the provisions of Article 13 of the Electronic Identification Regulation.”

“Furthermore, structurally significant digital identity providers should be legally designated as critical entities under the provisions of the Critical Entities Resilience Directive to ensure they receive the oversight and real-time threat intelligence necessary to strengthen global supply chain resilience.”

Romero-Moreno says legislation must also mandate the deployment of privacy enhancing technologies, or PETs, such as zero-knowledge proofs (ZKP) and synthetic biometric data.

Finally, the paper recommends establishing a binding transatlantic regulatory and technology coalition to “navigate the 12-month governance vacuum created by the AIA’s staggered enforcement timeline extending late into 2027. This alliance must mandate high-assurance standards – injection attack detection and zero-retention biometrics – as core financial risk-mitigation imperatives.”

Article Topics

biometrics  |  biometrics research  |  deepfake detection  |  identity assurance  |  identity governance  |  injection attack detection  |  presentation attack detection