Tax Notes logo

A Reexamination of Cashaw

Posted on June 26, 2023
Christopher Yan
Christopher Yan
Benjamin Alarie
Benjamin Alarie

Benjamin Alarie is the Osler Chair in Business Law at the University of Toronto and the CEO of Blue J Legal Inc., and Christopher Yan is a senior legal research associate at Blue J Legal.

In this article, Alarie and Yan revisit Blue J’s prediction in Cashaw after the Fifth Circuit affirmed the Tax Court’s holding that the taxpayer was liable for trust fund recovery penalties.

Copyright 2023 Benjamin Alarie and Christopher Yan.
All rights reserved.

In this month’s issue of Blue J Predicts, we revisit the intriguing case of Cashaw.1 This case focused on Pamela Cashaw, an administrator for a financially distressed hospital, who the Tax Court determined was personally liable for a $173,000 trust fund recovery penalty (TFRP) under section 6672. The fundamental issue was whether Cashaw was a responsible person who had willfully failed to fulfill her legal obligation to remit employee payroll tax withholdings, also known as trust fund taxes, thereby warranting the imposition of a TFRP.

In our November article we used Blue J’s TFRP prediction algorithm to assess the likely outcome of an appeal.2 Blue J predicted with 86 percent confidence that the Fifth Circuit would affirm the Tax Court’s decision if it endorsed the Tax Court’s findings of fact. Our analysis also considered alternate scenarios and examined the circumstances under which the Tax Court’s decision might be reversed. That exercise highlighted the importance of scrutinizing various factors in the case.

Now fast-forward to May 31. The Fifth Circuit affirmed the Tax Court’s decision, reiterating Cashaw’s liability for the trust fund recovery penalties.3 That validated Blue J’s prediction that Cashaw was a responsible person who had willfully neglected to pay. The result from Blue J’s TFRP predictive model, trained on the facts of more than 375 court opinions from 1956 to 2022 (trained up to the date of the prediction), underscores the transformative power of machine learning in conducting nuanced legal analyses.

As we reexamine Cashaw, we not only delve into the context of TFRP and the factors in the appeal but also reflect on the role our machine-learning-based prediction played in the analysis of this case. The alignment of machine-identified factors with those factors that have been decisive on appeal generates support for broader discussions on the future of artificial intelligence in legal decision-making.4

Notably, our Blue J Predicts contribution in May strayed from tradition to compare large language models like ChatGPT (versions GPT-3.5 and GPT-4) with our proprietary Ask Blue J, a chatbot specifically trained to accelerate tax research.5 This comparative analysis was intended to outline the benefits, potential concerns, and areas for improvement the deployment of generative AI to respond to complex tax law queries. Ask Blue J, which is engineered exclusively for tax law, generates responses derived from a meticulously curated tax document database.

As we return to our established tradition of revisiting predictions on pending cases this month, we embark on a fresh journey — investigating how generative AI can enhance the research process. While generative AI may not yet be mature enough to replace machine learning predictions in the litigation spotlight, we see its potential in aiding our research.

To illustrate its use, we have provided sample legal queries and responses below. Also, we have dedicated a section to dissecting the current limitations and challenges of using large language models in litigation. We provide insight into the strategies we are enthusiastically exploring to overcome these issues in the development of Ask Blue J. As part of our ongoing research, we discuss some promising ways generative AI can support our work, along with Blue J’s plans to tackle AI’s inherent challenges, furthering our journey into the radical improvement of tax law research with new technology.

TFRP and Section 6672

Section 6672 of the Internal Revenue Code, central to the TFRP, is invoked when employers neglect their responsibilities concerning the withholding of federal income taxes and Social Security contributions. Employers must retain these withheld amounts “in trust for the United States.” Should an employer default and that failure is deemed willful, section 6672(a) imposes a penalty equal to the total unpaid tax amount on any person required to collect, account for, or pay the withheld taxes.

Although that information can be found through traditional research methods, a practitioner unfamiliar with this area could become quickly acquainted with the fundamental aspects of section 6672 by posing a basic query to a tax-trained generative AI chatbot like Ask Blue J. Here is an example of a query and response:

User prompt: What is the section for the trust fund recovery penalty?

Ask Blue J: The section for trust fund recovery penalty is section 6672 of the Internal Revenue Code. This section imposes a penalty equal to the unpaid taxes on any person required to collect, truthfully account for, and pay over the withheld taxes, who willfully fails to do so. The liability is joint and several, and the Service is entitled to retain all amounts collected from responsible persons in satisfaction of their joint and several liability until such time that the Service’s right to retain the funds collected is established, either due to the expiration of the statutory period for commencing a refund suit, or if suit is filed, upon final adjudication of the action.6

As demonstrated, Ask Blue J correctly identifies the relevant section of the IRC and provides an overview of the TFRP. The summary, enriched by references to various field service advisory memoranda, offers insight into the joint and several nature of the liability. Ask Blue J can quickly help practitioners familiarize themselves with complex legal concepts and provide them with layered understanding, saving valuable time and effort in preliminary research.

TFRP liability is established by applying a two-part legal test. The first step involves determining whether an individual qualifies as “responsible” under section 6672. That determination is founded on the individual’s status, duty, and authority, as opposed to their understanding of the entity’s financial status. As delineated in Barnett by the Fifth Circuit,7 the factors relevant to this determination include the individual’s (1) status as an officer or member of the board of directors; (2) ownership of a substantial amount of stock in the company; (3) role in managing the day-to-day operations; (4) authority to hire or fire employees; (5) authority over funds disbursement and payment to creditors; and (6) authority to sign company checks.

These factors clarify whether an individual, by virtue of their position, had the “effective power” to ensure tax payments based on their actual authority or if they could have exerted “substantial” influence regarding those decisions.

Determining whether someone is a responsible person under this section is paramount to establishing liability. A practitioner engaged in initial research can use Ask Blue J to gain familiarity with the meaning of the term responsible person:

User prompt: What are some examples of who is a responsible person for section 6672?

Ask Blue J: Some examples of who is a responsible person for section 6672 include officers or employees of a corporation, members or employees of a partnership, and trustees who have responsibility, knowledge, and authority over the decision to pay or not to pay taxes. The determination of who is a responsible person is dependent on the facts of the case. In cases involving corporations and partnerships, the Service may collect the withholding taxes from individuals who have control over the financial affairs of the corporation or partnership. Examples of responsible persons may include officers or employees of a corporation who are responsible for collecting and paying over taxes, as well as individuals who have the authority to make financial decisions for the corporation. However, the specific determination of who is a responsible person would depend on the facts and circumstances of each case.8

Here, Ask Blue J provides examples of individuals who may be considered responsible persons under section 6672, with examples of responsibilities that the individuals would have in the organization, while reinforcing that the specific determination still depends on facts and circumstances. A review of the sources shows that Ask Blue J considered a Supreme Court case in preparing its examples of responsible persons.

If a person is deemed responsible, the analysis proceeds to the next step — determining whether their failure was willful. Willfulness in this context necessitates a voluntary, conscious, and intentional act of prioritizing other creditors over the United States. Importantly, proving willfulness doesn’t require evidence of a malicious motive or intent but typically involves evidence of other creditors being paid and the responsible person being aware of outstanding withholding taxes owed to the United States.

That provides another opportunity to use generative AI:

User prompt: How is it determined whether a responsible person was willful for section 6672?

Ask Blue J: Willfulness for section 6672 is determined by whether the responsible person’s actions were voluntary, conscious, and intentional. There does not need to be an intent to deprive or defraud the government of taxes to be paid over. A good faith misunderstanding of the law or a good faith misunderstanding that one is not violating the law negates willfulness. Therefore, to determine whether a responsible person was willful for section 6672, it must be determined whether their actions were voluntary, conscious, and intentional. If there was a good faith misunderstanding of the law or a good faith misunderstanding that one is not violating the law, then willfulness may be negated.9

The absence of any requirement of an intent to deprive or defraud the government in determining willfulness foreshadows challenging circumstances for Cashaw. While she presents a sympathetic narrative, the evidence suggests that her decision to prioritize some obligations above her duty to pay over taxes wasn’t driven by a misunderstanding of her obligations but was a calculated decision. As discussed below, the successful application of the reasonable cause and good-faith defenses to the TFRP has typically involved reliance on a tax adviser who either assured the taxpayer that the taxes had been paid or that they need not be paid. That bolsters the perception that Cashaw’s situation was a product of prioritization rather than misunderstanding.

In some circuits, a reasonable cause defense to the TFRP is acknowledged. However, no clear consensus exists on the precise definition of reasonable cause or the conditions needed to validate it. Typically, when the defense is successful, it involves the taxpayer’s reliance on the advice of a tax professional who assured that trust fund taxes were properly paid or were not required to be paid.

It is worth noting that the evidentiary factors for determining responsibility and willfulness can interrelate. Evidence bolstering the responsibility analysis may also have relevance to the willfulness analysis. For instance, an individual possessing the status, duty, and authority of a CFO is generally considered responsible and would typically be expected to be conscious of and responsive to financial difficulties impinging on payments to creditors. Thus, while someone’s title and role might prominently influence the responsibility analysis, these factors can also contribute to the willfulness analysis.

Recap: The Facts and Dispute in Cashaw

Cashaw served Riverside General Hospital (Riverside) in Houston, Texas, for more than 34 years, beginning as a pharmacist and later assuming senior administrative duties. Following the sudden removal of the chief administrator because of Medicare fraud allegations, Cashaw was appointed temporary chief administrator in October 2012. Her responsibilities in this role included overseeing Riverside’s operations, attending board meetings, and signing checks after reviewing hospital expenses. Despite her role, Cashaw argued that her authority was limited and shared, emphasizing her nonexclusive check-signing capability and lack of officer, board member, or shareholder status.

Cashaw’s tenure as administrator coincided with a period of financial distress for Riverside, which faced a withdrawal of Medicaid and Medicare funding and legal action from a major creditor, Dixon Financial Services. Despite the ensuing Rule 11 agreements designed to manage those issues, Riverside failed to meet some of its federal tax obligations in 2013 and 2014.

Constrained by those financial challenges and obligations under the Rule 11 agreements, Cashaw found herself in a difficult position. The government contended that this resulted in Cashaw attempting to prioritize payments among staff, vendors, and private creditors that provided “essential patient care services,” sometimes against the terms of the Rule 11 agreements. That included refusing to sign checks that, in her view, didn’t align with patient services.

Cashaw’s authority at Riverside was a major point of contention at trial. The government argued that she was a responsible person because of her chief executive administrative role. Cashaw disagreed, highlighting that she did not manage day-to-day operations or have hiring or firing authority. She claimed that her role was merely to approve pre-filled checks for staff payments and essential vendors. However, she recounted refusing to sign a check when she disagreed with its intended use.

At trial, Cashaw argued that her awareness of Riverside’s unpaid withholding taxes didn’t equate to willful nonpayment. Instead, she said she prioritized essential patient services, citing her obligations to patients under state law and the strictures of the Rule 11 agreements, which complicated Riverside’s financial situation.

Finally, Cashaw stated that she believed in the commitment of Gerry Hilliard, Riverside’s “financial consultant and de facto controller,” to pay the taxes. Hilliard had assured Cashaw of this commitment and formed a special committee to secure a loan for Riverside’s tax liabilities. That, combined with a previous successful cash injection, led Cashaw to believe further funds would be sourced to settle the tax debt.

Recap of the Tax Court Opinion in Cashaw

The Tax Court determined that Cashaw was a responsible person who acted willfully in not remitting full employment tax amounts owed to the IRS from Riverside. The court noted that by prioritizing payments to staff, vendors, and creditors when Riverside’s financial resources were strained, Cashaw displayed the level of control required to have significant influence over whether the funds would be allocated to federal taxes.

The Tax Court found Cashaw willfully noncompliant because she didn’t dispute her awareness of Riverside’s failure to fully pay its trust fund taxes. Despite knowing that, Cashaw continued to prioritize vendors and creditors over the federal government. The court wasn’t persuaded by Cashaw’s argument that the hospital’s funds were encumbered by other legal obligations. Also, the court dismissed her claim that she was acting under the direction of Hilliard when paying other creditors.

Cashaw’s potential personal exposure to civil sanctions and criminal penalties under Texas law, should she fail to pay other creditors, was also raised. However, the court countered that even if that were proven, it wouldn’t supersede section 6672. Although Cashaw didn’t present a reasonable cause defense, the court determined it was unavailable to her because she knew of Riverside’s unpaid federal employment tax liabilities and still paid other creditors instead.

While the Tax Court acknowledged the difficult position Cashaw was in, it clarified that its role was to uphold the law, not equity. Sympathy for Cashaw’s predicament couldn’t override the fact that she failed to collect and remit withheld funds. Even though her intentions may have been to prioritize essential patient care services, the court held that that did not make her failure to pay any less willful. Thus, her actions violated section 6672, resulting in her being held liable for the TFRP.

Revisiting Blue J’s Prediction

In the first article on the assessment of Cashaw, we used Blue J’s machine-learning model to ascertain the likelihood of Cashaw being held liable for the TFRP. Given the Tax Court’s findings, Blue J predicted liability with 86 percent confidence. If Cashaw had held positions of greater authority — such as director, officer, or shareholder — or had control over hiring or firing, the confidence level in the TFRP prediction would have exceeded 95 percent.

In our previous analysis, we noted that Cashaw’s emphasis on the absence of three Barnett factors — the individual’s role in managing the day-to-day operations; the individual’s authority over funds disbursement and payment to creditors; and the individual’s authority to sign company checks — didn’t alter the algorithm’s prediction because those elements were already factored out of the initial scenario.

Likewise, Cashaw’s claim that she held only a limited, nonexclusive ability to approve expenses and sign checks did not sway the prediction. Interestingly, the argument advanced by the taxpayer on this front appeared to be in tension with the Brown opinion, which explained that an individual doesn’t need to be the final authority on creditor payments to be held liable for a TFRP.10

Our analysis further highlighted discrepancies between scenarios 1 and 2 in terms of Cashaw’s control over payroll. While the Tax Court found that Cashaw had considerable payroll oversight and could prioritize staff payments, she portrayed herself as merely a rubber-stamper of pre-filled payroll checks, with no real control or responsibility over payroll or payroll taxes.

Cashaw’s reliance on Hilliard’s statements regarding the handling of trust fund taxes emerged as a substantial point in her defense. However, the TFRP data set did not indicate any significant correlation between relying on another person’s statements and the likelihood of avoiding a TFRP — unless, that is, the statement was made by a lawyer or accountant assuring appropriate handling of the taxes, or if it was made under conditions amounting to fraudulent or negligent misrepresentation. Neither of those circumstances appeared to be present in Cashaw’s case.

Cashaw’s argument that her preexisting duties as a pharmacist conflicted with her duty to handle trust fund taxes was treated as a novel issue. The closest similar case in Blue J’s data set involved another creditor asserting a legal right to be paid before the federal government.

We further explored several hypothetical scenarios to challenge the Blue J prediction as follows:

  • Scenario 2a: Hilliard, Riverside’s financial consultant (and de facto controller), made fraudulent or negligent misrepresentations to her regarding the trust fund taxes being addressed appropriately.

  • Scenario 2b: Cashaw tried to change payroll personnel or procedures upon learning of the tax delinquency.

  • Scenario 2c: Fifth Circuit prioritizes her preexisting duties as a pharmacist and hospital administrator over the federal government’s claim to trust fund taxes.

  • Scenario 3: Cashaw reviewed and approved payroll, but the factors in scenarios 2a, 2b, and 2c are also present.

Scenario Testing in Cashaw


Reviewed and Approved Payroll?

Accountant’s Fraudulent or Negligent Misrepresentations?

Changed or Attempted to Change Personnel or Procedures Around Payroll?

Duties to Patients Take Priority, Analogous to Creditor’s Priority Lien?

Predicted Outcome

Confidence in Predicted Outcome

1 (Tax Court)










































Those scenarios introduced new elements such as fraudulent or negligent misrepresentation by Hilliard, Cashaw’s attempts to change payroll procedures, and the possible precedence of Cashaw’s preexisting duties over federal tax claims. The results of these scenario tests presented a more optimistic view for Cashaw, suggesting that under some conditions, she might have a reasonable shot at showing that she was not willful or that she had reasonable cause for her actions.

Blue J’s Predictions and Fifth Circuit Ruling

In this section, we draw parallels between the data-driven predictions made in Blue J’s November 2022 publication and the recent May decision of the Fifth Circuit on appeal. This intersection of prediction and outcome is enlightening, and we show how our analysis foreshadowed the court’s holding against Cashaw, affirming her status as a responsible person and her willful failure to pay trust fund taxes.

First, the court relied on the six factors laid out in Barnett to gauge whether Cashaw was a responsible person under section 6672(a). Cashaw argued for her lack of responsibility based on the first five factors, but the court identified three pertinent factors. Cashaw’s assertions revolved around her minimal leadership role in the company — specifically that she lacked the status of an officer or board member and neither managed Riverside’s daily operations nor held hiring or firing authority. However, her authority to sign checks for Riverside and her role as chief administrator overseeing the hospital’s functions, including its payrolls, led the court to deem her a responsible person.

Figure 1 from our November article highlights how the interpretation of responsibility often transcends an individual’s title and role, extending to their knowledge and response in a willfulness analysis. To bolster that idea, a review of case law from 1956 to 2022 revealed that those who review and approve payroll — regardless of their official role or signing authority — are found responsible 90 percent of the time. Meanwhile those who possess formal authority to disburse funds as a director or officer but never actually do so are found responsible only about 60 percent of the time.

That is yet another area in which generative AI can assist in understanding the law:

User prompt: Can someone be a responsible person under the trust fund recovery penalty even if they are not a director or officer?

Ask Blue J: Yes, someone can be a responsible person under the trust fund recovery penalty even if they are not a director or officer. Section 6672 of the Internal Revenue Code imposes personal liability on any person required to collect, truthfully account for, and pay over withheld income and social security taxes who willfully fails to do so. Each “responsible person” is severally liable for the entire unpaid trust fund tax liability, including interest and penalties. Potentially responsible persons for purposes of the trust fund recovery penalty include employees, stockholders, surety lenders, and other persons or entities outside the delinquent employer corporation. When determining who is a responsible party, the IRS looks at who signs and files the employment tax returns and who makes the federal tax deposits. The determination of who is the person under a duty to collect, account for, and pay over employment and withholding taxes for wages paid to employees depends on the facts of the case.11

Ask Blue J correctly indicates that an individual can be a responsible person even if they are not a director or officer. It first sets out the framework for a responsible person by indicating that any person required to collect, truthfully account for, and pay taxes could be liable under section 6672. It draws its answers from several chief counsel advice memoranda and even cited a revenue ruling that considers whether a volunteer member of a board of trustees can be considered a responsible person under section 6672.

Cashaw further contested the willfulness of her failure to pay the taxes, professing she was assured the taxes would be paid and hence insisting that her failure was not willful. The court, however, made a stark comparison of her case with Gustin,12 marking the clear difference in Cashaw’s approach: She had not made “every reasonable effort” to ensure the taxes were paid, and her belief that tax payment had been arranged did not exempt her from personal liability under section 6672. The fact that Cashaw was aware of Riverside’s tax delinquency yet chose to prioritize essential patient services put her in violation of her duty as a responsible person.

Reflecting on Scenario 2 in our previous publication, we postulated that even if Cashaw’s situation mirrored Gustin insofar as Cashaw didn’t oversee payroll, Blue J’s TFRP algorithm would still predict a 61 percent likelihood that she would be liable. Moreover, the commentary from our previous article mirrors the Fifth Circuit’s analysis that distinguishes the extraordinary efforts taken in Gustin from the actions of Cashaw, who accepted Hilliard’s promises without contacting the IRS or conducting further inquiries. Our decision-search feature highlighted that the Gustin case paralleled the Cashaw situation regarding influence and control, but not in aspects of knowledge or response to the delinquency. Therefore, because of the significant differences in the fact patterns, the Blue J TFRP algorithm didn’t identify Gustin as similar enough to Cashaw to predict an absence of TFRP liability.

Also, Cashaw attempted to use her duty to her patients as a shield against the charge of willfulness. Despite the difficult position she faced as chief administrator, the circuit court did not find this argument compelling. The willfulness inquiry does not depend on motive. Once aware of the tax liability, the court held that a responsible person has a duty to ensure that taxes are paid before authorizing payments to vendors or employees.

Echoing our earlier analysis, we had considered a scenario in which a creditor is defending a legal right to be paid before the IRS. Comparing that with Cashaw’s situation, while we acknowledged a small chance of success, our tool had predicted, at a 59 percent confidence level, that she wouldn’t escape liability for the TFRP. Predictably, the court rejected her argument that her duties to her patients took precedence, regardless of the standards of care set by the state of Texas.

Finally, Cashaw asserted a reasonable cause defense for her failure to pay the trust fund taxes. The court, however, clarified that the reasonable cause defense is rarely successful, and does not extend to a responsible person who consciously chooses to pay other creditors before the government despite knowledge of due withholding taxes. Our prior article, encompassing all then-available TFRP cases, found that there was no consensus among the circuits regarding the validity of a reasonable cause defense. Further, our data analytics indicated that this defense succeeded less than 1 percent of the time. Successful defenses typically involved reliance on the advice of a tax professional that the trust fund taxes were being appropriately paid or did not need to be paid.

In summary, the Fifth Circuit’s findings weighed heavily in favor of the IRS. The decision affirmed that Cashaw was both a responsible person and acted willfully, echoing the predictive analysis presented in our November article. Our use of data analytics tools made it possible to surface those insights and accurately predict the intersection of our analysis with the court’s holding.

Integrating Generative AI in Litigation

As the field of AI continues to evolve, integrating generative AI into tax litigation practice is an intriguing prospect. However, several challenges require careful navigation and thoughtful consideration.

A stark example of the pitfalls can be seen in the case of a lawyer who used ChatGPT to prepare court filings in a lawsuit against an airline last month; ChatGPT fabricated fictitious court cases, which the lawyer cited in his filing.13 Although the lawyer believed the cases to be real, he nevertheless faces potential repercussions for the erroneous filing. That situation underscores the risk of generative AI “hallucinating” — which is the term used for when ChatGPT presents erroneous information as fact, in this case creating nonexistent scenarios or cases.

May’s Blue J Predicts article also delved into this potential issue, illustrating various ways that ChatGPT could hallucinate. This phenomenon has directed our approach to integrate generative AI into our processes, with an explicit priority to minimize these hallucinations. Thus, with Ask Blue J we are adopting a rigorous and thoughtful strategy for case inclusion, taking the utmost care to prevent hallucination of fictional cases.

The current version of the Ask Blue J prototype incorporates only Supreme Court cases, a decision borne out of the need for caution and close observation as we add new sources to the model. This choice naturally introduces limitations in its use, such as restricting the available precedent to leading cases that are typically discussed in IRS documents. Despite the restrictions, we firmly believe in the need for diligence, given the inherent complexity of legal cases.

Court case opinions are fundamentally different from other legal documents. They often reference outdated laws, contain extensive discussions on the interpretation of law by other courts, and can be overturned, distinguished, or otherwise affected by other opinions and changes in the law. A cavalier approach to including cases in a database could misguide the AI and introduce errors. Our commitment, therefore, is to ensure the accuracy and relevance of included cases. The development team at Ask Blue J is working diligently on introducing more cases into our database, while maintaining our guarantee of authenticity; all cases surfaced will be real, not invented by the AI chatbot.

To continue elevating our platform, we are now working meticulously to include other appellate cases into our database. It is crucial for us to curate these cases judiciously, ensuring relevance and utility for practitioners. It is imperative that we avoid inundating our system with technically correct but potentially unhelpful information, which might hamper the user experience when adding case law to the database.

Another challenge that generative AI solutions face in the tax context is the phenomenon of circuit splits. A prime example of this can be seen in our article discussing the applicability of reasonable cause as a defense in TFRP cases. The Fifth Circuit acknowledged the application of the test by other circuits but rejected adopting those tests because Cashaw didn’t demonstrate how these tests would operate within its existing case law on section 6672 liability. This situation underlines the necessity of considering each circuit’s unique jurisprudence when generating an Ask Blue J answer.

Moreover, we recognize there is a marked difference between searching for cases and asking a question from a database of tax documents — the former is instructional, while the latter is exploratory. This distinction speaks to our active exploration of combining these two approaches. The short-term objective is to enhance search and retrieval capabilities using large language models through the inclusion of case law text. Looking ahead, our aspiration is to build a multistep process that can retrieve, digest, and provide insights from the most relevant cases, assisting in tax outcome prediction with the benefit of generative AI.


Cashaw is a reminder of the unyielding nature of tax law. Despite the highly sympathetic circumstance of a hospital administrator striving to prioritize patient care amidst financial constraints, the Fifth Circuit applied the law unflinchingly. Even when faced with a party’s compelling and sympathetic conflicting duties, the Fifth Circuit has shown their resolve to hold a responsible person liable, emphasizing the importance of tax responsibilities.

Since this column’s inception in 2021, Blue J Predicts has consistently leveraged the power of data analytics and machine learning to offer nuanced predictions about tax law outcomes. Our journey commenced with the deployment of various Blue J Tax predictive modules. As we evolve, we are broadening our technological toolkit by investigating the potential of generative AI to augment these predictions. This exploration is not a departure from our reliance on machine learning, but rather an enhancement, adding another layer of analysis.

Generative AI’s application in dispute resolution, while still in its early stages, already shows immense promise. Today, it aids in the preliminary stages of research, and we anticipate that the incorporation of case law into the Ask Blue J dataset will substantially increase its practical value. With the ability to sift through relevant precedent and provide answers tailored to specific circuits, this tool promises a more comprehensive understanding of complex tax law issues, as demonstrated by our analysis of Cashaw.

Reflecting on Cashaw this month therefore highlights the continuity of our predictive success and our commitment to harnessing emerging technologies to assist our analysis. This ongoing exploration transcends mere technological innovation; it is fundamentally about equipping tax law practice with enhanced predictive power, adaptability, and versatility. It is about enabling professionals to traverse the complexities of tax law with a more informed and data-driven approach.

In every case we have analyzed, including the 15 tax appeals for which an opinion has been filed since 2021, Blue J Tax’s predictive algorithms have successfully forecasted the outcome. However, it is essential to reiterate that these tools don’t replace the legal expertise, judgment, and intuitive understanding that legal professionals provide. It is the combination of human insight and data-driven predictions that promises to unlock a new frontier in the realm of legal decision-making.


1 Cashaw v. Commissioner, T.C. Memo. 2021-123.

2 Benjamin Alarie and Ann Velez, “Cashaw: Conflicting Duties and the Trust Fund Recovery Penalty,” Tax Notes Federal, Nov. 28, 2022, p. 1257.

3 Cashaw v. Commissioner, No. 22-60024 (5th Cir. 2023), aff’g T.C. Memo. 2021-123.

4 Abdi Aidid and Alarie, The Legal Singularity: How Artificial Intelligence Can Make Law Radically Better (2023).

5 Alarie et al., “The Rise of Generative AI in Tax Research,” Tax Notes Federal, May 29, 2023, p. 1509.

6 Sources cited by Ask Blue J: Rev. Proc. 2005-34, 2005-1 C.B. 1233; IRS Publication 963 (11/2014); FSA 200106010; ECC 201323022; Rev. Proc. 2014-63, 2014-53 IRB 1; ECC 201811015; ECC 200908045; ECC 200912030; FSA 199904032; and FSA 199921007.

7 Barnett v. IRS, 988 F.2d 1449 (5th Cir. 1993).

8 Sources cited by Ask Blue J: FAA 20032901F; Rev. Rul. 84-83, 1984-1 C.B. 264; Slodov v. United States, 436 U.S. 238 (1978); and ILM 200026024.

9 Sources cited by Ask Blue J: Rev. Rul. 54-158, 1954-1 C.B. 247; and FSA 1992-92.

10 Brown v. United States, 464 F.2d 590, 591 n.1 (5th Cir. 1972).

11 Sources cited by Ask Blue J: ILM 200838027; Rev. Rul. 84-83, 1984-1 C.B. 264; ILM 200916024; ILM 200026024; ILM 200532046.

12 Gustin v. United States, 876 F.2d 485 (5th Cir. 1989).

13 Benjamin Weiser, “Here’s What Happens When Your Lawyer Uses ChatGPT,” The New York Times, May 27, 2023.


Copy RID