Anti-Corruption and Integrity Outlook 2024

9. Artificial intelligence

Introduction

The evolution in artificial intelligence (AI) is one of the most significant technological developments in recent years, and it presents many opportunities for anti-corruption and integrity actors. AI, including machine learning, can transform how anti-corruption and integrity actors identify, predict and mitigate fraud and corruption risks at a scale that was not possible with traditional rules-based approaches. This potential has a range of applications, from detecting irregularities in public spending or procurement processes, to expediting the review of asset declarations or legislative texts. AI, including the latest developments in generative AI and large language models, offer exciting new possibilities for analysing unstructured data, automating activities and creating efficiencies. Among other benefits, this allows auditors, investigators and analysts to focus more on issues and tasks that require expert knowledge.

However, the effectiveness and efficiency of any AI or analytics-driven approach for mitigating fraud and corruption risks not only relies on the availability and quality of data, but also on the skills of those responsible for it. AI and advanced analytics are not replacements for human judgement and expertise. And while AI can be a powerful ally for integrity actors, it also is a potential tool for others to perpetuate fraud and corruption with greater efficiency and on a larger scale.

This chapter explores these issues, and sets out how:

• AI is a tool to fight fraud, corruption and foreign bribery.

• AI has weaknesses and can be used to enable corrupt activities.

AI as a tool to fight fraud, corruption and foreign bribery

AI to prevent and detect public sector fraud and corruption

In the wake of the COVID-19 pandemic and in response to the cost-of-living crisis, many governments face both old and new fraud risks, some at unprecedented levels, linked to spending on relief and recovery, as well as public procurement. These are particularly high-risk areas, where any fraud and corruption can divert taxpayers’ money away from essential support for individuals and businesses. AI offers opportunities for governments to assess risks and predict likely fraud or corruption in ways that would have been previously impossible, or prohibitively resource intensive (Box ‎9.1). AI has made it easier for investigators and auditors to prioritise finite resources and improve the focus of data collection requirements, thereby saving taxpayer money (OECD, 2021[71]). It can also be used to improve asset declaration verification, to help verification agencies focus on high-risk declarations and reduce discretionary decision making. These systems flag declarations based on external signals (such as media reports, complaints from external parties, or referrals from other public authorities). Analytical software then compares flagged declarations with previous declarations from the filer, or from similar filers, to verify submissions and find patterns which can be used to develop new red flags for future verifications (Kotlyar and Pop, 2021[72]).

Box ‎9.1. Existing uses of AI to uphold integrity in OECD countries

Canada Employment Insurance (EI) Sickness Programme

AI supports ongoing investigation into the abuses of the EI benefit programme by focusing on identifying fictitious doctor’s notes. Once such allegedly fictious notes are discovered, they are associated with EI benefits to select cases for investigation. The project uses transcriptions and images and employs a variety of AI enabled technologies to extract relevant information from them. For example, Natural Language Processing (NLP) is applied to the transcripts to extract details about doctors. Optical character recognition (OCR) is used to extract that information from medical images while network analysis helps to identify claimants related to the known or newly identified cases of fraud.

UK HMRC

AI supports a number of activities including identifying risks on some large-scale transactional services, such as repayment claims for Value Added Tax (VAT) and Income Tax Self Assessment. UK HMRC are using analytics to help identify risks that need attention and building case packages that are passed to teams of investigators. AI also works well to assimilate large amounts of data – this is a newer implementation, important for compliance casework where HMRC are using AI alongside other tools like geo-mapping.

Swedish Tax Agency (STA)

The STA uses artificial intelligence to improve audit and risk-evaluation in relation to free-text information in the 150 000 handwritten income tax returns it receives every year. AI is used to interpret and convert handwritten text into digital text and to classify the text into one of around 60 subject categories. The text is interpreted using a deep-learning model that has been developed and trained by the STA. The benefits of this automated interpretation are that information reaches the right competence faster and that quantifying and analysing the content of free-text information is more effective, meaning identifying and processing potential fraud and corruption cases is quicker and more targeted.

Source: (International Public Sector Fraud Forum, 2020[73]; OECD, 2023[74]).

AI can also help make efficiencies and improvements in administrative systems which can reduce corruption. It can improve the efficiency of public services and administrative processes through reducing costs, enhancing productivity, and improving the quality of services and products offered to the public. But it can also reduce the opportunities for corruption among public office holders. Complex and unclear rules, red tape and inefficiency, and public office holders feeling disempowered by the system are among the main drivers of corruption in the public sector (U4, n.d.[75]). AI can address these issues by simplifying administrative procedures or reducing public office holders’ use of discretionary powers. For example, rules and regulations can be encoded as computer code and integrated into automated systems. These systems can then ensure that rules are applied consistently and fairly, reducing the risk of human error or the abuse of discretion, and improving the efficiency of the process (Parycek, Schmid and Novak, 2023[76]).

Given their ability to manage large databases, AI-driven methodologies can help to detect fraud and corruption in parts of the public sector dealing with large-scale transactional services, including healthcare, customs, infrastructure, procurement, or tax (Box ‎9.2). In particular, AI can help public institutions recognise abnormal interactions with their systems, and identify suspicious and potentially fraudulent activity through the use of, for instance, image recognition software or risk modelling (International Public Sector Fraud Forum, 2020[73]). It can also reduce the incidence of false positives, helping to sift otherwise valid interactions with public processes and services out of flagged cases (International Public Sector Fraud Forum, 2020[73]; OECD, 2021[8]).

Box ‎9.2. Tax administrations’ use of Artificial Intelligence to combat fraud and corruption

AI is increasing tax administrations’ abilities to combat fraud and corruption by improving risk management processes, supporting taxpayer compliance and reducing taxpayer burdens. Certain innovations are freeing up resources for targeted use in complex or higher-risk cases.

Compliance risk management

More countries are using AI, including machine learning, to understand current and future compliance risks, sharpen risk management, and develop intervention actions. 82.5% of countries surveyed in the OECD’s Tax Administration 2023 report are already using AI for risk assessments and detecting fraud in their tax systems, or are in the implementation phase for future use. One example includes incorporating AI auto-encoder techniques to compress high dimensional data and a short-term memory neural network to extract information from longitudinal (sequence) financial and economic entity data, as well as economic entities’ structure, which are used as predictors to enhance non-compliance prediction. Elsewhere, AI is being used to identify high-risk fraudulent tax deduction claims, such as an individual claiming deductions for travel expenses to work whilst lacking income from employment, and automatically refusing the claim where it cannot be substantiated. Overall, AI is allowing tax administrations to take a more preventative approach to fraud risk management. By seeking to intervene at earlier stages in taxpayer processes, tax administrations can prevent non-compliance, including fraud and corruption, rather than having to uncover it after tax returns have been filed.

Supporting taxpayer compliance and reducing taxpayer burden

AI is also being used to improve compliance by making it easier for taxpayers to engage with tax services and to meet their tax obligations. AI is helping facilitate the drive towards self-service, on a real-time and 24/7 basis, and opening new service options for tax administrations that allow more ‘compliance-by-design’ approaches. For example, AI is being used to deliver more sophisticated virtual or digital assistance services which can cope with more complex questions from taxpayers and provide more personalised answers. It is also being used to simplify customer interaction with tax systems, through automating the categorisation and routing of enquiries to the appropriate experts for action. And AI has been used to provide instant approval on tax assessments, reducing average assessment times and the duration and complexity of taxpayers’ interaction with tax authorities. These innovations have encouraged taxpayer interaction with tax systems, reducing disengagement, fraud and error, and allowed officials to focus on more complex cases with the potential for higher tax return.

However, risks and challenges in the use of AI to fight fraud and corruption in tax systems remain:

• Maintaining the security and privacy of information stored and used by AI tools, and only using data in a way compliant with data protection rules.

• Retaining levels of transparency in AI tools which maintain public trust in tax administrations, whose effectiveness depends on taxpayers’ consent and compliance.

• Ensuring legislative frameworks remain fit for purpose in the context of technological developments, not least in terms of data use and storage.

• Nurturing public engagement and trust in AI tools, especially in relation to the rollout of digital and virtual assistance tools.

• Ensuring access to third party data and communication between different systems to enable AI to learn and function optimally.

Source: (OECD, 2023[74]).

AI increasingly relevant in the fight against foreign bribery, both from the law enforcement and corporate perspectives

AI tools can be used at various stages of a foreign bribery enforcement action. As anti-money laundering regulations impose reporting obligations from financial institutions and Designated Non-Financial Businesses and Professions (DNFBPs), member countries have adopted AI tools to sort, connect, and prioritise data in suspicious transactions reports (STRs). One of the main benefits of using AI tools in this context is the identification of potentially relevant STRs connected to serious crime without the need to exhaustively describe all characteristic attributes of various typologies in the form of mathematical rules. Instead, these rules are automatically derived from data training (OECD, 2021[77]).

AI can also have multiple applications in foreign bribery investigations. AI can decipher code words and colloquialisms used in communications related to the bribery transaction, and expedite the preparation of investigation documentation. For example, the United Kingdom Serious Fraud Office reported using an AI tool in the context of a foreign bribery investigation to sort through and remove from evidence documents subject to attorney-client privilege. Scanning as many as 600 000 documents a day, AI reduced by 80% the pool of privileged material needing to be reviewed by independent counsel. In addition to reducing the review process time from two years to a few months, the use of AI also allowed for a more accurate and consistent review of the evidence (OECD, 2021[77]).

Companies can also use AI and data analytics to enhance their anti-corruption compliance and security and risk-management systems. AI-driven approaches in this area can for instance improve the monitoring of transactions, help identify and flag irregularities such as suspect payments, detect and prevent employee misconduct, and enable predictive analyses of corruption risks (Business at OECD, n.d.[78]).

AI weaknesses and enabling corrupt activities

Poor underlying data quality results in limitations to the use of AI

The use of AI can help integrity actors to efficiently prevent and detect corruption and fraud by drawing insights from large and complex datasets that would have been impossible to analyse otherwise. However, poor data quality can limit or undermine these efforts, potentially resulting in wasted resources or scepticism about the benefits of leveraging AI. Poor data quality has implications for a range of anti-corruption and anti-fraud activities. For instance, the pre-processing of data used for conducting fraud and corruption risk assessments, including assessing and addressing reliability issues, can be even more time-consuming than conducting the “analytics” itself (OECD, 2019[79]). Similarly, many integrity actors face the routine challenge of managing missing or error-prone data linked to a variety of critical data sources, such as registries for asset declarations or lobbying disclosures, as well as public procurement data.

Moreover, when the data used to train models are unreliable and incomplete, existing assumptions and biases can be perpetuated and even exacerbated (Adam and Fazekas, 2018[80]). These problems can arise due to pre-existing societal bias in the data, incomplete data, small sample sizes, errors in the definition of variables, or the omission or inclusion of flawed variables or proxies (OECD, 2019[9]; OECD, 2023[81]). The use of synthetic data (artificially generated data) for training AI models attempts to overcome some of the quality issues inherent in many complex datasets (Lee, 2024[82]), but there is still no full-proof measure to mitigate these risks. Flawed data matching and the use of error prone algorithms can have profound consequences (The Royal Commission into the Robodebt Scheme, 2023[83]).

Issues concerning data quality and AI can also exacerbate existing challenges around trust in public institutions and the auditability of decision making. AI tools can be seen as ‘black box’ systems, taking an input and producing an output while the process in between is neither visible nor easy to interpret. There is therefore a risk that the public may find it difficult to understand how and by whom decisions in public institutions are being made, with unintended impacts to the integrity and transparency of the process. Public institutions may find it hard to provide meaningful explanations of those AI processes, especially when security issues or intellectual property rights prevent them from doing so (International Public Sector Fraud Forum, 2020[73]). For many integrity actors, these challenges related to the interpretability and explainability of results can undermine the very principles they are meant to uphold, like transparency and accountability in public decision making.

In addition, it can be difficult for audit and integrity bodies to audit AI systems that are making decisions based on poorly trained AI models, or decisions experimentally or intuitively based on big data, and to understand how particular AI tools work. Where auditors do not have the correct access level or expertise, it can be difficult to verify that AI systems are functioning as intended and the necessary risk assessment and treatment mechanisms are in place (OECD, 2023[81]). The challenges associated with explaining and auditing AI systems could make it harder for governments to nurture trust in public decision making and for the public to be confident policymaking is effective and serving the public interest (OECD, 2019[9]; OECD, 2023[81]).

The increasing use of AI by corrupt actors

Some OECD countries are increasingly concerned that individuals and organisations engaging in fraud are using ever more sophisticated methods to analyse large amounts of data and exploit vulnerabilities in the public sector (International Public Sector Fraud Forum, 2020[73]). Many of the techniques are not new, such as social engineering or phishing attacks, but AI—generative AI and LLMs, in particular—exacerbate the risks. For instance, threat actors can create fraudulent documentation and AI-generated synthetic identities with greater ease and efficiency than ever before, which can be used to bypass Know Your Customer procedures or claim eligibility for public benefits. These efforts can compromise the integrity of, for instance, taxpayer-funded relief programmes related to COVID-19 or the cost-of-living crisis, diverting funds away from those that need it.

AI can also be used as a tool to influence political debate and policymaking. It can be used, legitimately, to lobby public office holders by preparing analysis of draft legislation, amendments and consultation documents, for stakeholder mapping and building lobbying strategies, or for drafting submissions and position papers to decision makers in public institutions (NY Times, 2023[84]; WFD and POPVOX, 2023[85]). These methods can help lobbyists develop more effective input to policymaking, enabling them to better represent their interests to decision makers, and could level the playing field for public engagement opportunities (WFD and POPVOX, 2023[85]). However, where these techniques are misused, to provide biased or false information or overrepresent certain viewpoints, AI can more effectively influence policymakers away from the public interest and in favour of specific private interests or even those of foreign powers (Sanders and Schneier, 2023[86]; Nay, 2023[87]). In addition, AI’s ability to generate ‘deep fake’ content makes it harder to separate what is true from what is not. Recent examples demonstrate how deep fake content can be used to influence political campaigns, in a way which misrepresents opposing campaigns and illegitimately influences the course of debate in democracies (WFD and POPVOX, 2023[85]; Murphy, 2024[88]). Although the use of AI in influencing activity may not be problematic in itself, many countries’ lobbying safeguards, having been developed before widespread availability of these tools, may not be mitigating the risks of AI in lobbying. For instance, many countries do not have standards of conduct for lobbying activity which mitigate the risk of false interests being represented, let alone on the scale possible through use of AI. There is also clear scope for countries to develop their guidance to public office holders to help them assess the reliability of information presented to them (OECD, 2021[19]).

Overall, AI is offering OECD countries a range of opportunities to uphold public sector integrity more effectively, and it will do so increasingly in years to come. AI has fundamentally changed how public authorities assess and manage integrity and corruption risk in public institutions. However, countries should understand the limitations and risks associated with the use of AI in the public sector. Limitations are especially prevalent where countries are struggling to collect the data needed for AI to work. And countries should also consider how well prepared their lobbying frameworks are for managing AI-based influencing activity, especially in terms of their standards for use of AI in lobbying and their guidance to office holders on dealing with lobbying activity.