von Patrick Wellens

The Association of Certified Fraud Examiners (ACFE) mentions in their annual Report to the Nation that organizations lose between 5% and 6% of their annual revenue to fraud. The typical fraudster is male, aged between 31 and 50, while individuals over 50 are responsible for the highest median losses. The ACFE also developed the “fraud tree“ and the Cressey Fraud Triangle (comprising opportunity, incentive/pressure, and rationalization), a widely used framework among investigators and compliance officers to understand why individuals commit fraud.

With the widespread use of Artificial Intelligence, the question is whether fraud losses as a percentage of revenue will increase, whether the fraudster profile is likely to change and whether the fraud triangle framework remains a valid model to explain fraudulent behavior. What can compliance officers and investigators do better in preventing, detecting and investigating AI-enabled fraud?

AI and the fraud triangle

AI dramatically expands the opportunity for fraud. AI enables fraudsters to effortlessly generate falsified travel and expense receipts (hotel invoices, restaurant bills, taxi receipts etc.), false or fake supplier invoices, manipulated images (to support insurance fraud), and fabricated documents (company records, contracts, emails etc.). In addition, AI allows them to generate deepfake voice/video impersonating other people.

The arguments to rationalize fraud are also shifting with AI. Traditionally, fraudsters justified their actions with arguments such as “I worked really hard – I deserve this”, “upper management is doing it as well”, or “I was treated unfairly”. With AI, the rationalization becomes delegated to technology (e.g., “AI did it,” “I just executed what the system suggested,” “no one is directly harmed,” “it is just an algorithm” etc.).

AI does not change the pressure for fraud, however it lowers the effort and skill required to act fraudulently. Considering that, an increase in fraud incidents can be expected, unless companies strengthen their preventive and detective efforts.

AI and the fraudster profile

Before Artificial Intelligence, fraudsters were employees with university degrees, long tenure within the organization and often had substantial expertise in accounting, IT or operational processes.

The classic profile of a fraudster according to ACFE required access and skill, fraudsters were personally involved in the execution of fraud, fraud schemes developed gradually, and often behavioral red flags surfaced. Artificial Intelligence fundamentally disrupts this profile.

With AI, a much broader population of employees can engage in fraud, even without specialized skills.  The natural language model of AI removes technical barriers to commit fraud. External actors can impersonate insiders and therefore no longer need to be long-tenured employees.

Adjustments in fraud prevention, detection and investigation

Fraud prevention

Compliance officers in companies can implement the following measures to be more effective in the prevention of AI-related fraud:

• Compliance personnel should train employees and managers to create awareness on AI-driven fraud schemes. The best prevention remains the ability to recognize such schemes.
• Strong governance of AI tools.
• Logging of AI-generated content.
• Traditional fraud prevention relied heavily on segregation of duties, manual approval, and static thresholds. These methods are still important, however, for fraud prevention tools to be effective they must be adaptive, they must identify patterns and subtle anomalies in large datasets that indicate fraudulent behavior.
• To avoid the fake personification on voice/video approvals, strong authentication/verification controls should be implemented.

Fraud detection

Compliance officers in companies can implement the following measures to be more effective in the detection of AI-related fraud:

• Most fraud is detected via anomalies, not (static) thresholds. Therefore, AI models should be used to detect unusual (timing) transactions.
• Annual compliance monitoring or yearly audits are obsolete and must be replaced by real-time or near real-time monitoring.
• Anonymized datasets of T&E expenses, supplier invoices, and purchase orders can be uploaded to AI models and reviewed for anomalies or questionable transactions.

Fraud investigation

Compliance officers and/or investigators in companies can implement the following actions to be more effective in the investigation of AI-related fraud:

• Investigators increasingly need a strong understanding of AI models and must collaborate with IT, cybersecurity and data science teams. The future fraud investigator evolves from a traditional auditor into a hybrid forensic analyst and technologist.
• AI tools can be used in various stages of investigation. Instead of the traditional keyword-search approach applied on ten thousand structured and unstructured documents, investigators can use AI prompts to identify relevant evidence more efficiently.
• AI models can be used to prepare interview questions, visualize timelines of facts, conduct network analysis, etc.
• Investigators must verify authenticity of audio/video, distinguish between human and AI-generated content. Analyzing metadata as evidence and preserving AI model outputs becomes an important skill for fraud investigators.

Reshape strategies to combat fraud

The widespread use of Artificial Intelligence makes it much easier for employees to commit fraud. Fake invoices, receipts, documents, videos, etc. can be generated with minimal effort. As a result, fraud risks are likely to increase unless companies adapt their preventive, detective and investigative approach to AI fraud. Even though AI lowers the effort and skill for employees to act fraudulently, companies can also use Artificial Intelligence in numerous ways to become more effective in detecting and investigating fraud.

• AI Governance, logging /monitoring of AI generated content and awareness training are the best fraud prevention actions that can be taken.

• Increased focus on strengthened authentication controls is needed to prevent fake personification on voice/video materials.

• Fraud detection must be changed from static thresholds, annual compliance monitoring, or yearly audits to real-time monitoring of unusual transactions.

• AI models can significantly increase the effectiveness of investigations. AI models used in review of emails, network analysis, and creation of initial interview questions significantly reduces investigation time and costs.

Der nächste Beitrag auf dem Blog Economic Crime erscheint  am 27. April 2026

Autor: Patrick Wellens

Patrick Wellens is working as a Global Compliance business partner for a division of a multinational pharma company, based in Zurich, Switzerland. He is the Chair of Ethics and Compliance Switzerland, co-chair of the working groups “life science” and “anti-corruption” and Chief Compliance Officer of the Association of Corporate Investigators. He has almost 30 years’ experience developing effective governance, risk, ethics and compliance programs and helping companies respond to crises, collaborating with prosecutors, compliance monitors, and law firms.