In the modern digital banking landscape, it is essential for financial institutions (FIs) to adopt effective strategies to protect against the evolving threat of check fraud. As fraud techniques become more sophisticate, ranging from counterfeit checks to altered payee names; detecting these illicit activities can be increasingly challenging. Fortunately, by harnessing advanced technologies like Artificial Intelligence (AI), FIs can proactively streamline their check fraud detection processes. This approach not only enhances security but also significantly mitigates associated risks, ultimately fostering a safer banking environment for both financial institutions and their customers/members.

The impact of check fraud in the US has been substantial:

• The U.S. Treasury Department reports a staggering 385% increase in check fraud nationwide since the onset of the pandemic.
• An overwhelming 90% of U.S. FIs surveyed have confirmed that check fraud is escalating, despite a decline in check volume.
• In 2023, a significant 47% of organizations faced fraud involving paper checks.

The above findings clearly demonstrate that check fraud poses a serious and escalating threat to both financial institutions and their customers/members in the United States. This alarming trend persists even as the overall use of checks declines. It is imperative for stakeholders to acknowledge this risk and take action to protect themselves.

The pandemic accelerated the shift to remote banking, which, while convenient, exposed financial institutions to new vulnerabilities. Many scammers targeted mailboxes and postal carriers, intercepting genuine checks and altering or “washing” them to change names or amounts. These modified checks are often traded on the dark web or used to create counterfeit versions.

To combat the risk of check fraud, FIs must look out for several behavioral and document-based red flags:

1. Behavioral Indicators

• Rushed Transactions: Customers/members rushing for immediate check clearance might be attempting to bypass standard scrutiny.
• Vague Details: Customers/members not willing to provide clear information about a check’s origin might look suspicious.
• Geographical Irregularities: Stolen checks often reappear at different branches, sometimes in other states.

2. Documentary Clues

• Altered Checks: Inconsistencies in signature, fonts, or ink might be signs of tampering.
• Check Stock Variations: Fraudsters using technology to create checks might not perfectly replicate the original check stock or layout.
• Inconsistent Signatures: Authentic signatures have distinct characteristics, such as loops and letter connections. Forged signatures may show irregular “bleed points” or heavy strokes.

Traditional fraud detection systems often struggle to keep pace with the complexity and intricacies of modern-day check frauds. Fraudsters are increasingly employing advanced tactics, like counterfeit checks, altered payee names, and check washing, thus exploiting gaps in conventional fraud detection systems. To minimize these gaps, FIs need to use cutting-edge technologies that are advanced enough to combat these financial frauds at their disposal.

This is where Artificial Intelligence (AI)—specifically Generative AI (GenAI) and Agentic AI can make a significant difference by providing advanced solutions to modern challenges like:

1. Image Quality and Manipulation
One of the key challenges in detecting check fraud is dealing with poor-quality or distorted check images, which fraudsters may manipulate to hide their alterations.

• Agentic AI can assist in improving the quality of such images even before investigation of fraud begins. This system can automatically enhance image resolution, applying techniques like sharpening and contrast adjustment to ensure that check images are clear enough for analysis. Agentic AI can also self-flag blurred or forged images and forward them for more intense scrutiny.
• Large Language Model (LLM)-assisted data augmentation can help train the fraud detection models to identify fraudulent alterations in checks even when the images are of low quality or taken under less-than-ideal lighting conditions.

2. Complex Data Matching:
Check fraud often involves mismatched or inconsistent data—such as names, amounts, or dates that need to be cross-checked across internal and external systems.

• LLMs are proficient in language comprehension and can handle such unstructured data. This makes it feasible to cross-check names, amounts, and dates across systems, even if the data is inconsistently formatted or stored.
• Agentic AI can continuously update its models based on real-time customer/member transaction data, allowing it to adapt to evolving fraud tactics and reduce false positives.

3. Detecting Synthetic Identities and Stolen Accounts:
Synthetic identities are created by combining real and fake data to appear legitimate, which is also a threat to check fraud.

• LLMs can recognize subtle language and pattern indicators of synthetic identities, such as unusual name constructions or implausible combinations of customer/member details. They can help flag potential synthetic identities by recognizing the synthetic patterns in large data sets, even if they look legitimate at first glance.
• Agentic AI can verify & identity details across external sources in real-time, ensuring whether the identity information aligns with typical identity profiles and transaction patterns.

4. Behavioral Analysis Limits:
Check fraud also involves certain anomalies in typical customer/member behavior, which are too subtle to be noticed.

• Agentic AI can track and analyze customers’/members’ behavioral trends in real-time, adjusting to new customers/members or infrequent check users. Over time, the system builds a profile to understand what’s “normal” for each customer/member, minimizing false positives even without extensive historical data.
• LLMs can provide reasoning behind flagged anomalies using natural language explanations. This makes it easier for analysts to interpret why certain checks are flagged as suspicious.

5. Real-Time Detection and Processing:
Real-time detection of check fraud might be challenging due to the need to process large volumes of checks quickly. Fraudsters might alter the checks technically, making it challenging to detect fraud immediately.

• Agentic AI can split detection tasks into smaller parts, such as verifying signatures, amounts, and payee information independently and in parallel, to speed up processing without sacrificing accuracy. This makes it feasible to detect fraud in real time, even for high volumes of transactions.
• LLMs can provide faster insights and predictive scores for potential fraud, reducing the need for human intervention. LLMs enable fraud detection systems to prioritize checks with higher risk, ensuring that lower-risk transactions are processed without delay by assessing risk factors in real-time.

6. AI Bias and Transparency:
AI-driven fraud detection systems can inadvertently develop biases due to training data, resulting in false positives or missed fraud. Additionally, many AI models operate as “black boxes,” making it difficult to interpret or explain their decision-making processes, which raises concerns about fairness and trust.

• LLMs can generate human-readable reasons for why certain transactions are flagged as fraudulent. This promotes transparency and minimizes potential biases by clarifying decision pathways, making it easier to spot and correct biased patterns in AI models.
• Agentic AI systems can continuously monitor detection decisions for any sign of bias and autonomously adjust parameters or flag areas of potential bias, creating more equitable fraud detection outcomes.

7. Adaptation to New Fraud Tactics:
Fraudsters constantly refine their check fraud tactics to create increasingly convincing counterfeit checks. It is imperative for fraud detection systems to evolve simultaneously with innovative fraud tactics to protect the FIs against future losses.

• Agentic AI systems can autonomously identify new fraud patterns as they emerge, rapidly adapting to novel tactics like check washing. These agents can upgrade detection systems without waiting for scheduled updates by constantly retraining and adjusting themselves.
• LLMs can generalize from existing fraud examples to understand variations of known fraud tactics, helping them detect new tactics that are slight variations of previous ones. For instance, an LLM can detect different check-washing strategies by analyzing the fundamental concepts behind check-washing.

8. Inter-FIs Data Sharing and Privacy Concerns:
Sharing check fraud data across FIs can mitigate the risk of financial losses but raises significant privacy and security concerns. Balancing fraud prevention by avoiding privacy breaches of customer/member personal details requires advanced technical solutions.

• FIs can simultaneously share data insights without revealing actual customer/member data by using LLMs trained on federated learning frameworks. This safeguards customers’/members’ privacy while gaining insights from other institutions. This approach allows each FI to take advantage of shared learning without compromising customer/member privacy.
• Agentic AI can independently detect and report fraud patterns across multiple institutions without sharing customer/member-specific data. It can also help alert FIs to wider fraud schemes by focusing on aggregated patterns and trends rather than individual transactions.

In collaboration with external data sources like the Financial Crimes Enforcement Network (FinCEN), GenAI and Agentic AI improve compliance with anti-fraud regulations while respecting privacy. They facilitate shared fraud insights without compromising on customer/member data via federated learning, thus enhancing inter-FI cooperation.

The advent of Generative AI and Agentic AI signals a transformative shift in check fraud detection. These advanced technologies empower financial institutions (FIs) to identify fraud with unparalleled speed and precision, significantly lowering operational costs and risks. By harnessing these innovations, FIs can enhance transaction safety, boost customer/member trust, and stay ahead of evolving fraud tactics, ultimately future-proofing their services.

– by Sharad Gupta
   Vice President, Automation & AI