Algorithmic Accountability and Anti-Corruption Governance: A Framework for Detecting Substandard and Falsified Medical Products

Abstract

The proliferation of substandard and falsified (SF) medical products represents a critical failure of pharmaceutical governance that disproportionately harms populations in low- and middle-income countries. Criminological research has long documented the structural conditions that enable pharmaceutical corruption, from corporate crime in drug testing and manufacturing to the transnational networks that traffic falsified medicines across regulatory boundaries. While international anti-corruption instruments, notably the United Nations Convention Against Corruption (UNCAC) and the Council of Europe MEDICRIME Convention, provide legal frameworks to combat pharmaceutical corruption, significant enforcement gaps persist. Algorithmic decision-making systems are increasingly deployed to detect SF products within pharmaceutical supply chains, yet these systems operate largely outside established accountability frameworks and may themselves reproduce the structural biases that characterise global pharmaceutical markets. This article proposes an Algorithm Accountability Benchmark (AAB) designed to evaluate the transparency, fairness, and accountability of algorithms used in pharmaceutical product detection and regulatory enforcement. Drawing on the theoretical frameworks of criminogenic asymmetries, global anomie and dysnomie, and the concept of practices that are “lawful but awful,” the paper argues that algorithmic governance tools must be assessed not only for technical performance but also for their capacity to either reinforce or disrupt the structural conditions that enable pharmaceutical corruption. The AAB integrates anti-corruption governance principles derived from UNCAC with algorithmic accountability standards from the European Union Artificial Intelligence Act, offering a comprehensive evaluation methodology across prevention, detection, reporting, and enforcement dimensions. The article contributes to the growing interdisciplinary literature linking anti-corruption law, algorithmic governance, and global health security.

1. Introduction

Substandard and falsified (SF) medical products constitute one of the most consequential forms of transnational crime, operating at the intersection of corruption, public health, and regulatory failure. The World Health Organization estimates that approximately 10.5% of medical products circulating in low- and middle-income countries (LMICs) are substandard or falsified, a figure derived from quality surveys encompassing over 48,000 samples across 88 countries (WHO 2017a, 2022). Recent systematic reviews suggest prevalence rates as high as 17–22% in certain therapeutic classes and regions (Ozawa et al. 2018; Zabala et al. 2022), while countries spend an estimated USD 30.5 billion annually on SF products (WHO 2024). Beyond economic costs, these products cause treatment failure, antimicrobial resistance (Cavany et al. 2023), and tens of thousands of preventable deaths each year; WHO-commissioned modelling estimates that between 72,430 and 169,271 childhood pneumonia deaths annually are attributable to SF antibiotics alone (WHO 2017a; Zabala et al. 2022). The emergence of falsified semaglutide (Ozempic) products across multiple WHO regions, prompting a global medical product alert in 2024, demonstrates that even tightly regulated high-income markets are vulnerable (WHO 2024).

The trade in SF medical products is not merely a regulatory or public health problem; it is fundamentally also a corruption problem. Four decades of criminological research have documented the structural conditions that enable pharmaceutical crime. Braithwaite’s (1984) pioneering study of corporate crime in the pharmaceutical industry, based on interviews with 131 senior executives across five countries, documented pervasive patterns of international bribery, fraud in drug safety testing, and criminal negligence in manufacturing. His findings remain disturbingly relevant. The subsequent analysis by Dukes et al. (2014) extended this work to demonstrate how corporate pharmaceutical crime had evolved in scope and sophistication, embedding itself within legitimate supply chain structures to a degree that makes detection and attribution of responsibility increasingly difficult. Pharmaceutical corruption—encompassing bribery in drug approval, procurement fraud, the falsification of manufacturing and testing records, and the deliberate distribution of substandard products—undermines patient access to effective medicines by diverting public resources for private gain and eroding trust in health systems (Kohler and Dimancesco 2020; Kohler and Ovtcharenko 2013; Kohler et al. 2016; Transparency International 2006; Vian 2020). Transparency International’s Global Corruption Report on health identified the pharmaceutical sector as one of the most corruption-prone segments of the global economy, with vulnerabilities at every stage of the value chain from product approval through procurement to dispensing (Transparency International 2006).

These dynamics are amplified by what Passas (1998, 2001) has termed “criminogenic asymmetries”: structural discrepancies and inequalities in law, regulation, enforcement capacity, and economic conditions across national jurisdictions that generate opportunities for illicit profit and motives to take advantage of such opportunities, while simultaneously undermining social controls. The persistence of these asymmetries reflects conditions of global anomie and dysnomie—the erosion or fragmentation of effective norms and regulatory controls in a context of deep political-economic inequalities—in which legality and illegality coexist uneasily, and actors may cause severe social harm while remaining formally compliant with fragmented regulatory regimes (Passas 2000, 2022).

In response, the international community has developed legal instruments aimed at combating pharmaceutical corruption and the trade in falsified products. The United Nations Convention Against Corruption (UNCAC), entered into force in 2005, provides a comprehensive framework addressing prevention, criminalisation, international cooperation, and asset recovery (UNODC 2004). The Council of Europe’s MEDICRIME Convention, which entered into force in 2016, represents the first binding international criminal law instrument specifically targeting the counterfeiting and falsification of medical products (Council of Europe 2011). At the European level, the Artificial Intelligence Act (EU AI Act), adopted in 2024, establishes a risk-based regulatory framework for algorithmic systems, classifying certain healthcare applications as high-risk and imposing requirements for transparency, accountability, and bias mitigation (European Parliament and Council of the European Union 2024b).

Simultaneously, algorithmic decision-making systems have been deployed with increasing frequency in pharmaceutical regulation and supply chain monitoring. These systems take several forms. Portable near-infrared (NIR) spectrometers coupled with machine learning classifiers are used as first-line field screening tools, enabling rapid, non-destructive identification of substandard and falsified medicines at points of entry and in community pharmacies; recent studies have demonstrated that such devices can discriminate between genuine and falsified antimalarial formulations with high sensitivity and specificity, even in resource-limited settings (Caillet et al. 2021; Vickers et al. 2018). Anomaly detection algorithms are applied to pharmaceutical procurement data to identify patterns indicative of bid-rigging, price manipulation, or suspicious supplier behaviour, the kind of large-scale pattern recognition across thousands of simultaneous transactions that would be impossible for human auditors (Schwalbe and Wahl 2020). Track-and-trace serialisation systems, mandated under the EU Falsified Medicines Directive and the US Drug Supply Chain Security Act (and the blockchain-based architectures proposed to support them), create tamper-evident digital records of pharmaceutical supply chains that algorithmic tools can monitor in real time (Zakari et al. 2022). INTERPOL’s Operation Pangea targets the online sale of illicit medicines; Operation Pangea XVII resulted in the shutdown of approximately 13,000 criminal-linked websites, social media pages, channels, and bots (INTERPOL 2025).

Yet despite these diverse applications, these algorithms largely operate outside any coherent accountability framework. They may encode biases reflecting the structural inequalities that characterise global pharmaceutical markets or may simply be poorly suited to the operational environments in which they are deployed. For example, NIR spectroscopy classifiers trained on spectral libraries from European manufacturers encounter fundamental accuracy problems when applied to the diverse product formulations found in LMIC markets, where frequent changes in materials sourcing and formulation by local manufacturers are common (Caillet et al. 2021; Vickers et al. 2018). As Obermeyer et al. (2019) demonstrated in the context of healthcare risk prediction, algorithmic systems can systematically disadvantage already-marginalised populations when the data on which they are trained reflects existing patterns of structural disadvantage.

This paper addresses a gap at the intersection of three fields that have developed largely in isolation: anti-corruption governance, algorithmic accountability, and pharmaceutical regulation. It proposes an Algorithm Accountability Benchmark (AAB) that integrates anti-corruption principles derived from UNCAC and the MEDICRIME Convention with the algorithmic governance standards established by the EU AI Act. Drawing on the theoretical frameworks of criminogenic asymmetries (Passas 1998, 2001), global anomie and dysnomie (Passas 2000, 2022), and the concept of “lawful but awful” practices (Passas 2005), and building on the empirical analysis of the social organisation of illicit medical-product markets (Passas, forthcoming), the paper argues that algorithmic systems can serve as instruments of anti-corruption governance, but only if they are subject to accountability mechanisms as rigorous as those demanded of the human institutions they supplement or replace.

2. Theoretical Framework: Criminogenic Asymmetries, Corporate Crime, and the Legal–Illegal Interface

Understanding how SF medical products emerge and persist in global pharmaceutical markets requires a theoretical framework that goes beyond individual criminal acts to address the structural conditions that enable them. Three interconnected bodies of criminological theory are essential to this analysis: the concept of criminogenic asymmetries, the study of corporate crime in the pharmaceutical industry, and the framework of global anomie and dysnomie.

2.1. Criminogenic Asymmetries and Global Anomie

The concept of criminogenic asymmetries, developed by Passas (1998, 2001), refers to structural discrepancies, mismatches, and inequalities in the spheres of politics, culture, technology, the economy, and the law that (1) generate or strengthen the demand for illegal goods and services, (2) create incentives for particular actors to engage in illegal practices, and (3) reduce the capacity of authorities to detect and control crime. In the pharmaceutical domain, these asymmetries manifest across multiple dimensions. Legal asymmetries arise from the vast differences in regulatory capacity and enforcement among national jurisdictions: a product that would be immediately flagged as substandard by the stringent regulatory authorities of high-income countries may circulate freely in markets where national medicines regulatory authorities lack the technological capacity, resources, or political independence to conduct effective quality monitoring (Passas, forthcoming; WHO 2017b). Economic asymmetries create demand-side drivers: more than two billion individuals in LMICs lack adequate access to essential medicines, and where populations cannot afford quality-assured products, parallel markets in cheaper, but potentially substandard or falsified, alternatives emerge (Wirtz et al. 2017; Yenet et al. 2023). In parts of Africa, the lowest-paid government employee may need to work roughly two weeks to afford a one-month supply of essential medicines, generating inelastic demand that unscrupulous suppliers exploit (Passas, forthcoming). Political asymmetries operate when corruption within regulatory agencies creates impunity for manufacturers and distributors of SF products, as procurement processes are captured by private interests (Kohler and Ovtcharenko 2013; Kohler et al. 2016).

These asymmetries are not isolated phenomena but expressions of deeper normative dislocations. The framework of global anomie and dysnomie (Passas 2000, 2022) captures the conditions under which effective norms and regulatory controls are eroded or fragmented in a context of intensified inequalities produced by globalisation. Anomie describes the breakdown or weakening of norms’ guiding power; dysnomie describes situations where laws and regulations are present but ineffective, inconsistently enforced, or applied in a discriminatory manner. In the pharmaceutical sector, dysnomie is pervasive: legal frameworks exist in most countries, yet the gap between their formal provisions and their practical enforcement creates environments in which legality and illegality coexist, and actors cause severe harm while remaining formally compliant with fragmented regulatory regimes. These conditions are particularly pronounced in the governance of algorithmic systems, where the very novelty of the technology means that regulatory norms are either absent, untested, or designed for different institutional contexts.

2.2. Corporate Crime and Regulatory Capitalism

The criminological study of pharmaceutical corporate crime provides essential empirical and conceptual foundations for understanding why algorithmic accountability matters. Braithwaite’s (1984) landmark investigation revealed that corporate crime in the pharmaceutical industry was not aberrational but structurally embedded. Bribery of government officials to expedite drug approvals, fraud in safety testing data, and criminally negligent manufacturing practices were documented across jurisdictions with widely varying regulatory environments. The study demonstrated that pharmaceutical companies systematically exploited regulatory asymmetries—testing drugs in countries with lower safety standards, seeking first approval in jurisdictions with lax requirements, and using the resulting approvals to leverage entry into more strictly regulated markets. Three decades later, Dukes et al. (2014) documented how these patterns had intensified under globalisation, with corporate pharmaceutical crime becoming increasingly transnational and supply chain structures increasingly complex, diffusing responsibility across multiple legal entities and jurisdictions in ways that obstruct accountability.

Braithwaite’s (2008) concept of regulatory capitalism (the idea that the contemporary economy is characterised not by deregulation but by the proliferation of regulatory institutions, standards, and mechanisms, often privatised or hybrid in form) is directly relevant to algorithmic governance. Algorithmic systems deployed in pharmaceutical supply chain monitoring are themselves regulatory instruments: they classify, flag, approve, and sanction. Yet they operate within institutional arrangements that Braithwaite would recognise as exhibiting the characteristic tensions of regulatory capitalism, where regulatory functions are delegated to private actors (algorithm developers, technology companies) whose incentive structures may diverge from public health objectives. The risk, as Passas and Nelken (1993) noted in the context of subsidy fraud, is that the “thin line between legitimate and criminal enterprises” becomes even thinner when algorithmic systems designed to detect fraud are themselves designed and operated by actors embedded in the commercial structures they are meant to regulate.

2.3. The Legal–Illegal Interface and “Lawful but Awful” Practices

The typology of the “legal–illegal interface” directs analytical attention to the porous boundary between legal and illegal practices in transnational commerce (Passas 2005, forthcoming). In the pharmaceutical sector, this interface is strikingly visible. Passas (forthcoming) has documented how illicit medical-product markets are hybrid systems: illicit actors do not operate in isolation but with the involvement (sometimes knowing, sometimes negligent) of legitimate manufacturers, distributors, customs officials, logistics companies, and health professionals. The 2025 WHO-UNODC report on contaminated medicines confirmed that intentional criminal conduct, including deliberate falsification of excipients and documentation, contributes directly to multiple contamination incidents (WHO and UNODC 2025).

These dynamics illustrate what Passas (2005) has described as “lawful but awful” practices: actions that may comply with the letter of existing regulations, but that exploit regulatory gaps, perpetuate inequalities, or undermine public health objectives. In pharmaceutical supply chains, such practices include outsourcing production to jurisdictions with weak oversight, reliance on paper-based verification for complex chemical inputs, and contractual fragmentation that diffuses responsibility across multiple legal entities. While such practices may not be illegal, their cumulative effect is to create systemic vulnerability that criminal actors readily exploit. The result is a form of harm production that is structurally embedded in global commerce rather than driven solely by overt criminal intent (Braithwaite 2008; Dukes et al. 2014; Passas 2005; Passas and Nelken 1993).

These theoretical tools are essential for evaluating the contribution of algorithmic systems to anti-corruption governance. The central question is not merely whether an algorithm performs accurately in a controlled testing environment, but whether it operates in a manner that addresses or reinforces the criminogenic asymmetries that generate pharmaceutical corruption in the first place. An accountability framework that fails to ask this question risks legitimating technological solutions that leave the structural drivers of corruption untouched, a form of institutional performance that obscures rather than addresses the underlying problem.

3. The Social Organisation of Illicit Medical-Product Markets

Effective algorithmic accountability in pharmaceutical governance requires an empirically grounded understanding of how illicit medical-product markets actually function. Recent research demonstrates that these markets are characterised by organisational features that present specific challenges for algorithmic detection and governance (Passas, forthcoming).

First, illicit pharmaceutical networks exhibit multi-layered participation and modular organisational structures, with specialised functional roles for sourcing, manufacturing, packaging, shipment, distribution, online marketing, and payment processing (Passas, forthcoming). This organisational modularity mirrors legitimate pharmaceutical operations while providing operational security through compartmentalisation. Algorithms designed to detect anomalous supply chain patterns must therefore be capable of identifying disruptions across functionally distinct segments of the chain, rather than relying on single-point indicators.

Second, the hybrid character of pharmaceutical supply chains—where illicit products move through legitimate ports, distributors, and pharmacies to conceal their origin—means that the distinction between legal and illegal transactions is often invisible at any individual node in the chain (Braithwaite 1984; Dukes et al. 2014; Passas, forthcoming). The 2008 heparin contamination crisis, in which a deliberately adulterated product passed through multiple legitimate commercial intermediaries before reaching patients, exemplifies how harm can travel along supply chains while responsibility becomes dispersed and difficult to assign (Passas, forthcoming). Algorithmic systems that evaluate transactions in isolation, without reconstructing supply chain pathways, will miss precisely the forms of fraud that cause the greatest harm.

Third, digital enablement has transformed the reach and operational security of illicit pharmaceutical markets. Online pharmacies, social media marketing, encrypted communications, dark web marketplaces, and small-parcel shipping create new distribution channels that evade traditional regulatory controls (Passas, forthcoming). Operation Pangea XVII (December 2024–May 2025) resulted in 769 arrests, the seizure of 50.4 million doses worth USD 65 million across 90 countries, and the shutdown of approximately 13,000 criminal-linked websites, and other online channels; these were the largest seizures and arrests in the operation’s 17-year history. Yet these impressive figures represent only a fraction of the total market (Passas, forthcoming; INTERPOL 2025). These dynamics underscore the need for algorithmic tools capable of operating across both physical and digital supply chain environments.

Fourth, crisis exploitation is a defining feature of illicit pharmaceutical markets. The COVID-19 pandemic demonstrated how quickly criminal networks adapt to exploit healthcare emergencies, particularly in contexts of problematic governance (Passas 2022). Surging demand for semaglutide for weight loss similarly created rapid falsification opportunities, with products containing insulin rather than semaglutide causing hospitalisations for severe hypoglycaemia (Passas, forthcoming). Algorithmic systems must therefore be designed for adaptive responsiveness to rapidly evolving market conditions, rather than relying on static detection models trained on historical data.

4. Anti-Corruption Legal Frameworks Applicable to Pharmaceutical Integrity

4.1. The United Nations Convention Against Corruption (UNCAC)

The UNCAC is the most comprehensive international anti-corruption instrument, ratified by over 190 states parties (UNODC 2004). Several of its provisions are directly applicable to the governance of pharmaceutical supply chains. Article 9 requires state parties to establish transparent and competitive procurement systems based on objective criteria, with provisions for the prevention of conflicts of interest. This requirement applies directly to the procurement of medical products, where corruption is known to inflate prices, distort product selection, and facilitate the entry of substandard goods (Kohler and Ovtcharenko 2013; UNODC 2012). Articles 7 and 8 address codes of conduct, merit-based recruitment, and financial disclosure for public officials, all of which bear on the integrity of national medicines regulatory authorities. Article 21 addresses bribery in the private sector, relevant to corruption at the dispensing stage where private sector actors dominate the pharmaceutical value chain.

However, implementation of UNCAC is uneven. Country reviews conducted under the Convention’s Implementation Review Mechanism show that its non-mandatory provisions, including those on bribery in the private sector and illicit enrichment, have been implemented less consistently than its mandatory provisions on the bribery of public officials (UNODC 2017). In the pharmaceutical sector specifically, evaluations of good governance initiatives such as the Medicines Transparency Alliance have found a persistent gap between identifying governance weaknesses and remedying them (Kohler and Ovtcharenko 2013). This inconsistency creates precisely the kind of legal asymmetry that Passas (1998) identified as criminogenic: private sector pharmaceutical dispensing, which occurs at the point closest to the patient, operates largely outside the mandatory provisions of the UNCAC, leaving decision-making vulnerable to corruption. Corruption is a cross-cutting enabler of the entire SF trade: evidence from Africa and the Sahel shows pharmaceutical company employees, civil servants, law-enforcement officers, and health-agency workers among those implicated in trafficking networks, making control especially difficult when those tasked with oversight are themselves complicit (Passas, forthcoming; UNODC 2023).

4.2. The MEDICRIME Convention

The Council of Europe’s MEDICRIME Convention represents the first binding international criminal law treaty specifically targeting the falsification of medical products and similar crimes (Council of Europe 2011). Entering into force on 1 January 2016, the Convention criminalises the manufacture, supply, and trafficking of falsified medical products, including medicines for human and veterinary use, medical devices, and their components and active ingredients. It is deliberately open to accession by non-Council of Europe member states, reflecting the inherently transnational character of pharmaceutical crime. As of the end of 2025, ratification has expanded to 23 states, including several non-European states such as Benin, Burkina Faso, Côte d’Ivoire, Guinea, Mali and Niger, with additional non-Council of Europe countries among the signatories (Council of Europe Treaty Office 2026; Passas, forthcoming).

The MEDICRIME Convention is notable for its three structural pillars: criminalisation, victim protection, and international cooperation. By establishing harmonised criminal offences, it directly addresses one of the central criminogenic asymmetries in pharmaceutical governance: the disparity in legal treatment of falsified medicines across national jurisdictions. In many countries, the falsification of medical products has historically been treated as a minor regulatory infraction rather than a serious criminal offence, resulting in weak deterrence (Hamilton et al. 2016; Passas, forthcoming). Conceptual and terminological inconsistencies compound these gaps: terms such as “substandard,” “falsified,” “counterfeit,” and “spurious” are used differently in public health, intellectual property, customs, and criminal law contexts, complicating cross-border cooperation and mutual legal assistance (Passas, forthcoming; WHO 2017b). The MEDICRIME Convention’s requirement that states parties criminalise falsification as a substantive criminal offence, with proportionate sanctions, represents a normative shift toward treating pharmaceutical falsification as what it is: a form of public health corruption. The MEDICRIME Committee’s Strategy 2024–2025 responds to the changing character of the trade by targeting not only falsification itself but also its root causes and its promotion, particularly through online media (Council of Europe 2024).

4.3. The EU Artificial Intelligence Act and Algorithmic Governance

The EU Artificial Intelligence Act, adopted in 2024, introduces a risk-based regulatory framework for AI systems that is directly relevant to algorithmic tools deployed in pharmaceutical regulation (European Parliament and Council of the European Union 2024b). The Act classifies many AI systems used in healthcare as high-risk, notably those that are—or are safety components of—medical devices subject to third-party conformity assessment, subjecting them to mandatory requirements for transparency, human oversight, data governance, and technical documentation (Solaiman and Malik 2025). Data governance provisions require that training datasets be relevant, sufficiently representative, and to the extent possible, free of errors and complete. This is a demanding standard given the well-documented data quality challenges in pharmaceutical supply chain monitoring (WHO 2017b, 2022). The accountability provisions require that deployers monitor the operation of high-risk AI systems and report serious incidents to relevant authorities. Together, these provisions establish a regulatory architecture that, if applied to pharmaceutical supply chain algorithms, would constitute a meaningful accountability framework, though one that currently applies only within the European Union’s jurisdiction, creating yet another asymmetry in global governance.

5. The Algorithm Accountability Benchmark (AAB): Design and Principles

The Algorithm Accountability Benchmark (AAB) proposed in this article is a comprehensive evaluation framework designed to assess whether algorithms deployed in pharmaceutical supply chain monitoring and SF product detection satisfy the normative requirements of anti-corruption governance. Unlike purely technical benchmarking exercises, the AAB explicitly incorporates the governance principles embedded in UNCAC, the MEDICRIME Convention, and the EU AI Act, while being informed by the criminological analysis of how illicit pharmaceutical markets actually function (Braithwaite 1984; Dukes et al. 2014; Passas 1998, forthcoming). The benchmark is structured around four interconnected evaluation dimensions: prevention, detection, reporting, and enforcement, mirroring the functional architecture of anti-corruption frameworks themselves.

5.1. Transparency as the Foundation of Accountability

The AAB is grounded in the principle that transparency is a precondition for meaningful accountability. UNCAC Article 10 requires state parties to take measures to enhance transparency in public administration. When algorithmic systems assume functions previously performed by human regulators, the transparency obligations that applied to human decision-makers must transfer to the algorithmic system. This means that the data on which the algorithm was trained, the features it considers, the weighting it assigns to different inputs, and the confidence intervals associated with its outputs must all be documented and accessible to relevant stakeholders (Buhmann et al. 2020; Floridi 2021).

The “black box” problem is not merely a technical inconvenience in this context; it is a governance failure. Braithwaite (1984) documented how opacity in pharmaceutical manufacturing and testing processes systematically enabled corporate crime by preventing regulators from verifying compliance claims. The same dynamic applies to algorithmic systems: if the decision to classify a medical product as substandard cannot be explained to the manufacturer, the regulator, or the affected patient population, then the algorithmic system fails the most basic requirement of due process. The AAB therefore requires that algorithms deployed in pharmaceutical regulation meet minimum interpretability thresholds, calibrated to the severity of the consequences of their outputs.

5.2. Fairness, Context Mismatch, and the Problem of Structural Bias

The fairness dimension of the AAB addresses the risk that algorithmic systems may be inaccurate, unfit for purpose, or systematically biased when deployed in pharmaceutical supply chain contexts that differ from those in which they were developed. This risk manifests in several distinct but related forms.

The most fundamental problem is context mismatch: an algorithm designed for one regulatory environment may be unfit for the operational realities of another. A NIR spectroscopy classifier trained on spectral libraries from European manufacturers, for example, encounters accuracy failures when applied to the diverse product formulations found in LMIC markets, where frequent unreported changes in materials sourcing and product formulation are common (Caillet et al. 2021; Vickers et al. 2018). A supply chain anomaly detection model calibrated to the serialised, tightly regulated distribution networks of the EU or the United States may produce false positives when applied to the informal distribution channels (community pharmacies, open markets, mobile vendors) through which medicines reach patients in much of sub-Saharan Africa and South Asia. These are not bias problems in the conventional sense. Rather, they are fitness-for-purpose failures arising from the deployment of algorithms in environments fundamentally different from those in which they were trained. The consequences include not only missed SF products but also false identifications that burden already-stretched regulatory agencies and may disrupt the legitimate medicine supply.

Structural bias compounds these context mismatch problems. Obermeyer et al. (2019) demonstrated that a widely used healthcare algorithm in the United States systematically assigned lower risk scores to Black patients than to equally sick White patients, because structural racism in the healthcare system meant that less money was spent on Black patients at any given level of illness. The implications for pharmaceutical supply chain algorithms are direct: if an algorithm is trained on data from regulatory systems that underreport SF products—whether due to limited testing capacity, corruption-related suppression of adverse findings, or the structural marginalisation of certain patient populations from pharmacovigilance systems—then the algorithm will systematically underestimate the prevalence of SF products precisely where the problem is most severe.

From the perspective of criminogenic asymmetries (Passas 1998), context mismatch and structural bias are both products of the same underlying dynamics: the structural disparities across jurisdictions that generate pharmaceutical corruption also produce the data gaps and operational divergences that make algorithmic detection unreliable. Information asymmetries among key stakeholders constitute a major criminogenic factor in their own right. Companies, regulators, law-enforcement agencies, and international organisations often hold fragmented and incomplete data on SF products, while institutional reluctance to share information—driven by concerns over commercial confidentiality, reputational damage, and unclear liability—creates informational advantages for illicit actors (Passas, forthcoming). The AAB addresses these challenges by requiring disaggregated performance and fairness auditing across jurisdictional and demographic contexts, with particular attention to both accuracy disparities and detection rate disparities between high-income and low-income settings (Rajkomar et al. 2018). Algorithms that perform well in aggregate but fail in specific deployment contexts do not satisfy the AAB’s requirements.

5.3. Accountability, Liability, and the Attribution of Responsibility

The accountability dimension of the AAB addresses the question of who bears responsibility when an algorithmic system produces an incorrect or harmful output. Braithwaite and colleagues (Braithwaite 1984; Dukes et al. 2014) demonstrated that one of the most persistent challenges in pharmaceutical regulation is the diffusion of responsibility across multiple actors in complex supply chains. The globalisation of pharmaceutical manufacturing has intensified this problem: the 2008 heparin crisis revealed layers of workshops, consolidators, and processors across China before reaching US manufacturers. This complexity obstructed accountability to the point that the perpetrators were never identified or prosecuted (Passas, forthcoming). When algorithmic systems mediate regulatory functions, accountability can become further diffused among algorithm developers, deployers, data providers, and the institutions that rely on algorithmic outputs for regulatory decisions.

Several concrete liability scenarios illustrate the governance challenges the AAB must address. First, consider a manufacturer whose product is falsely identified as substandard by an algorithmic screening system, leading to product seizure, market withdrawal, or reputational damage. Under traditional regulatory frameworks, the affected manufacturer would have recourse against the regulatory authority that acted on the classification. When the classification is generated by an algorithm, however, the question arises whether the manufacturer’s claim lies against the deploying regulator, the algorithm developer, or both. The AAB requires that algorithmic systems deployed in regulatory contexts be accompanied by clear documentation of the allocation of liability between developers and deployers, and that effective remedial mechanisms, including expedited review procedures for contested algorithmic classifications, be established before deployment.

Second, consider the inverse scenario: a patient is harmed because an algorithm failed to detect a falsified product that it should have identified. The question of whether the algorithm’s failure constitutes a breach of duty depends on the standard of care expected of the system, a standard that, given the relative novelty of these technologies, remains contested. The EU’s revised Product Liability Directive (2024/2853), which entered into force in December 2024 and applies to products placed on the market after 8 December 2026, takes a significant step by explicitly treating software and AI systems as “products” subject to strict liability, with a rebuttable presumption of defectiveness where a product fails to comply with mandatory safety requirements laid down in Union law, such as those of the AI Act (European Parliament and Council of the European Union 2024a, 2024b). For pharmaceutical supply chain algorithms, this means that a failure to comply with the AI Act’s data governance, transparency, or testing requirements could give rise to a presumption that the algorithmic system was defective, a powerful incentive for developers to comply with the AAB’s standards.

Third, the question of whether algorithm developers should be permitted to contractually exclude or limit their liability for algorithmic failures in pharmaceutical contexts requires careful attention. The public health stakes are qualitatively different from those in commercial contexts where contractual freedom may be appropriate: a pharmaceutical detection algorithm that fails to identify a falsified antimalarial may contribute to preventable deaths. The AAB proposes that contractual exclusions of liability should not be enforceable in high-risk pharmaceutical applications, consistent with the approach of both the EU AI Act’s mandatory requirements for high-risk systems and the MEDICRIME Convention’s emphasis on proportionate sanctions for pharmaceutical crime (Council of Europe 2011; European Parliament and Council of the European Union 2024b).

Fourth, when algorithmic failures trigger investigations, a tension arises between transparency and the protection of proprietary algorithmic methods. The AAB requires sufficient disclosure to permit meaningful accountability, while recognising that trade secrets may be legitimately protected. The principle the AAB adopts is that any aspect of an algorithmic system that contributes to a regulatory decision affecting public health must be disclosable to competent authorities upon request, even where broader public disclosure is not required. This is consistent with the EU AI Act’s provisions requiring that high-risk AI systems be accompanied by technical documentation sufficient for authorities to assess compliance (European Parliament and Council of the European Union 2024b), and with the principle in responsive regulation theory (Braithwaite 2008) that the severity of disclosure requirements should be calibrated to the severity of the consequences at stake.

The AAB draws on the EU AI Act’s allocation of responsibilities to establish clear accountability chains for pharmaceutical algorithms, consistent with UNCAC Article 12’s prescription for private sector accountability including adequate auditing standards and effective sanctions for non-compliance. Critically, the AAB requires that algorithmic systems incorporate human-in-the-loop oversight mechanisms. This insistence on meaningful human review reflects the insight from responsive regulation theory (Ayres and Braithwaite 1992; Braithwaite 2008) that effective governance requires calibrated institutional responses, not automated mechanical application of rules. This principle is especially important when algorithmic outputs may trigger criminal referrals, product recalls, or market restrictions with significant consequences for public health and individual rights.

6. Evaluation Methodology: The Four Dimensions of the AAB

6.1. Prevention

In the prevention dimension, the AAB evaluates whether algorithmic systems contribute to the structural prevention of corruption and the entry of SF products into supply chains. This dimension assesses whether algorithms support UNCAC Article 9 procurement transparency requirements by analysing procurement data for patterns indicative of bid-rigging, price manipulation, or preferential contract awards. It also evaluates whether supply chain monitoring algorithms are integrated with Good Manufacturing Practice compliance systems in a manner that enables the proactive identification of manufacturing sites at risk of producing substandard products (Passas, forthcoming; Wirtz et al. 2017). Crucially, the AAB’s prevention assessment examines whether the algorithm’s design accounts for the criminogenic asymmetries that drive pharmaceutical corruption; specifically, whether it is calibrated to the risk profiles of different jurisdictions, the supply–demand gaps that generate informal markets, and the governance deficits that create regulatory blind spots (Passas 1998, forthcoming), rather than applying uniform thresholds derived from high-income regulatory contexts.

6.2. Detection

The detection dimension evaluates both technical and governance performance. The AAB assesses sensitivity, specificity, and real-time processing capability, but supplements these with governance-oriented assessments: whether the algorithm’s training data is representative of diverse supply chain architectures and product types (Mackey and Nayyar 2017); whether it performs equitably across geographic regions, income levels, and therapeutic categories; and whether the decision boundary appropriately balances risks of false negatives (with potentially lethal consequences) against false positives (with economic and access consequences) (Hamilton et al. 2016; Ozawa et al. 2018). The practical illustration is instructive: portable NIR spectrometers can achieve high accuracy in discriminating genuine from falsified antimalarials in controlled settings, yet their performance degrades significantly when the spectral libraries on which they depend do not include the specific manufacturers supplying a given market (Caillet et al. 2021; Vickers et al. 2018). The modular, networked structure of illicit pharmaceutical operations documented in another study (Passas, forthcoming)—where different groups handle formulation, packaging, shipping, distribution, marketing, and payment processing—demands detection algorithms capable of correlating anomalies across functionally distinct supply chain segments rather than evaluating individual transactions in isolation.

6.3. Reporting

The reporting dimension assesses whether algorithmic outputs are translated into timely, accurate, and actionable information. The World Health Organization’s Global Surveillance and Monitoring System (WHO GSMS) has documented persistent underreporting of SF products, driven by limited regulatory capacity, fear of retaliation, and absent reporting infrastructure in many LMICs (WHO 2017b, 2022). Studies show that many clinicians are aware that SF medicines exist but lack knowledge about how to recognise them or which channels to use for reporting (Passas, forthcoming). Algorithms can potentially address some of these barriers by automating the detection-to-reporting pipeline. The AAB evaluates whether algorithmic reporting systems generate outputs comprehensible to non-technical decision-makers; integrate with existing national and international mechanisms, including the WHO GSMS; and incorporate protections against the suppression of findings by corrupt actors, for example, through automated alerts that transmit findings simultaneously to multiple oversight bodies.

6.4. Enforcement

The enforcement dimension evaluates whether algorithmic systems support effective enforcement responses consistent with the MEDICRIME Convention’s criminalisation requirements and UNCAC’s provisions on international cooperation and asset recovery. The AAB assesses whether algorithmic outputs provide evidentiary documentation sufficient for criminal prosecutions; whether chain of custody for algorithmically generated evidence is maintained; and whether algorithmic systems facilitate cross-border information sharing essential for combating transnational pharmaceutical crime. This dimension also evaluates proportionality: whether algorithmic systems enable regulators to distinguish between substandard products resulting from manufacturing failures and falsified products resulting from deliberate fraud, a distinction that Braithwaite (1984) identified as critical for calibrating appropriate regulatory and criminal justice responses (Council of Europe 2011; Nayyar et al. 2015).

7. Data Quality, Governance, and the Challenge of Asymmetric Information

The effectiveness of any algorithmic system is bounded by the quality, representativeness, and integrity of its data. In the pharmaceutical domain, data quality challenges are inseparable from the governance failures the AAB seeks to address. The WHO GSMS, the most comprehensive global database of SF product reports, is itself shaped by the criminogenic asymmetries described above: countries with the weakest regulatory capacity—and consequently the highest burden of SF products—are also the least likely to detect and report them (WHO 2017b, 2022). Laboratory capacity is a foundational but unevenly distributed component of medicine-quality assurance: in many LMICs, national quality-control laboratories are under-resourced, lack accreditation, or are entirely absent. The Gambia cough-syrup tragedy demonstrated how reliance on foreign laboratories delays confirmation of contamination, allowing harmful products to remain on the market (Passas, forthcoming).

These data limitations are compounded by the information asymmetries among stakeholders discussed in Section 5.2. The AAB addresses data quality through several mechanisms: requiring disclosure of training data provenance, scope, and limitations, consistent with EU AI Act data governance requirements (European Parliament and Council of the European Union 2024b); mandating sensitivity analyses across settings with different reporting rates; and requiring periodic retraining to reflect evolving falsification methods, a necessary safeguard against the adaptive behaviour of criminal networks that continuously modify their operations in response to detection technologies (Mackey and Nayyar 2017; Passas, forthcoming).

8. Discussion: Algorithms, Anti-Corruption Governance, and the Structural Conditions of Pharmaceutical Crime

The framework proposed in this paper rests on a claim that is both pragmatic and normatively significant: algorithmic systems deployed in pharmaceutical governance can function as instruments of anti-corruption, but only if they are embedded within accountability structures that match the ambitions of international anti-corruption law. Without such accountability, there is a risk that algorithmic governance becomes an example of “lawful but awful” practice (Passas 2005); technically permissible under existing regulations, yet operating in ways that undermine the public interest by creating an illusion of effective oversight while leaving the structural drivers of pharmaceutical corruption intact.

The empirical evidence on how illicit pharmaceutical markets function—as hybrid systems blending legal and illegal actors, exploiting regulatory fragmentation, and adapting rapidly to enforcement pressure and crisis conditions (Passas, forthcoming)—underscores both the potential and the limitations of algorithmic approaches. Well-designed algorithmic systems can address governance failures that traditional institutional mechanisms have struggled to remedy alone. Blockchain-based track-and-trace systems can create tamper-evident records, reducing opportunities for corrupt intermediaries to divert genuine products (Zakari et al. 2022). Machine learning models can analyse procurement data across thousands of transactions simultaneously, identifying patterns of manipulation invisible to human auditors (Schwalbe and Wahl 2020). Automated reporting can reduce dependence on human intermediaries vulnerable to corrupt pressures, enabling the kind of real-time, multi-jurisdictional information sharing that UNCAC’s international cooperation provisions envision.

Conversely, the same structural analysis highlights serious risks that go beyond the reproduction of existing biases. An algorithm deployed in a regulatory environment fundamentally different from the one in which it was developed may be simply inaccurate or unfit for the task: a supply chain anomaly detector trained on European serialised distribution data may produce systematic errors when applied to the informal, fragmented distribution networks found in much of sub-Saharan Africa, not because it embeds discriminatory biases, but because it was built for a different world. The consequences include false positives that burden already-stretched regulatory agencies, missed indicators that allow dangerous products to reach patients, and a host of downstream errors that erode institutional confidence in algorithmic tools. These context mismatch failures interact with, and are compounded by, structural bias (Obermeyer et al. 2019)—where training data reflect existing patterns of disadvantage—and by the data quality gaps that characterise the settings where SF products are most prevalent. The erosion of public trust in health systems and pharmaceutical actors (documented as both a consequence and a driver of pharmaceutical corruption; Passas, forthcoming; Transparency International 2006) adds urgency. If algorithmic systems are perceived as instruments of market power or foreign regulatory imposition rather than public health protection, they may further undermine the institutional legitimacy on which effective pharmaceutical governance depends.

The framework of responsive regulation (Ayres and Braithwaite 1992; Braithwaite 2008) offers a model for how algorithmic governance might be structured to address these challenges. Responsive regulation prescribes an enforcement pyramid in which regulators begin with cooperative, dialogue-based approaches and escalate to more coercive measures only when cooperation fails. Applied to algorithmic pharmaceutical governance, this suggests a tiered approach. Algorithmic flagging triggers human review and dialogue with the affected manufacturer; sustained non-compliance escalates to formal regulatory action; and evidence of deliberate falsification triggers criminal referral. The AAB’s insistence on human-in-the-loop oversight operationalises this principle, ensuring that algorithmic outputs inform rather than replace the calibrated institutional judgments that responsive regulation demands. The emerging EU liability framework, in which non-compliance with the AI Act can create a presumption of defect under the revised Product Liability Directive (European Parliament and Council of the European Union 2024a), provides additional incentive structures for developers and deployers to comply with accountability standards such as the AAB.

The AAB is designed to maximise potential benefits while mitigating governance risks. By requiring transparency, disaggregated performance and fairness auditing, and human-in-the-loop oversight, the benchmark ensures that algorithmic systems remain instruments of, rather than substitutes for, human governance. By requiring that algorithms be evaluated against the specific normative standards of anti-corruption law, rather than purely technical performance metrics, the AAB insists that technological tools be held to the same accountability standards as the institutional systems they augment.

9. Conclusions

The trade in substandard and falsified medical products is not simply a crime problem; it is a structural feature of the globalised economy (Passas, forthcoming). Its persistence reflects deeper political–economic inequalities, regulatory fragmentation, the hybrid character of pharmaceutical supply chains in which legal and illegal actors are intertwined, and the conditions of global anomie and dysnomie that characterise contemporary pharmaceutical governance (Passas 1998, 2000, 2022). International anti-corruption instruments, particularly UNCAC and the MEDICRIME Convention, provide normative and legal frameworks for addressing this challenge, but their implementation is constrained by the very criminogenic asymmetries they seek to remedy.

The Algorithm Accountability Benchmark proposed in this paper offers a framework for ensuring that algorithmic systems deployed in pharmaceutical governance serve the objectives of anti-corruption law. Grounded in the criminological analysis of pharmaceutical corporate crime (Braithwaite 1984; Dukes et al. 2014), the structural theory of criminogenic asymmetries (Passas 1998, 2001), and the empirical study of how illicit medical-product markets are socially organised (Passas, forthcoming), the AAB provides a structured methodology for evaluating whether pharmaceutical algorithms satisfy requirements of transparency, fairness, and accountability. Its integration of responsive regulation principles (Ayres and Braithwaite 1992; Braithwaite 2008) ensures that algorithmic governance is designed for calibrated, contextually appropriate institutional responses rather than mechanical enforcement.

Future research should focus on the empirical application of the AAB to specific algorithmic systems currently deployed in pharmaceutical supply chain monitoring, testing whether the benchmark’s evaluation criteria can be operationalised in practice and identifying areas where refinement is needed. Comparative analysis across jurisdictions with different regulatory capacities would be particularly valuable, as would the assessment of the data quality and context mismatch challenges that constrain algorithmic performance in the settings where SF products are most prevalent. The most durable solutions are those that confront the structural determinants of the problem—entrenched health inequities, chronically weak regulatory capacity, global price asymmetries, and the erosion of public trust—rather than only its manifestations (Passas, forthcoming). Ultimately, the goal is to ensure that the technological tools increasingly central to pharmaceutical governance promote transparency, accountability, and the protection of public health across diverse global contexts.