With the rapid increase of generative artificial intelligence (AI) in all aspects of our lives, the federal government has established overarching parameters to guide federal agencies in using AI ethically and responsibly. For example, Executive Order 14110, Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence (October 30, 2023) requires agencies to develop guidelines, standards, safeguards, and best practices for AI reliability, safety, and security, while promoting innovation and advancing equity.[1] Federal agencies, such as the IRS, are exploring ways to use AI to improve efficiency and enhance capacity. Specifically, the IRS is striving “to harness the benefits of Al while mitigating its risks.”[2] The IRS issued interim guidance effective May 20, 2024 to communicate AI governance and principles.[3] Recently the IRS issued a memorandum on Privacy Policy for AI, effective September 30, 2024, to outline privacy requirements and considerations for the IRS’s own use of AI that applies to IRS users, developers, and providers of AI.[4] These IRS policies are intended to ensure the responsible use of AI at the IRS and “to create trust in the use of AI through responsible AI practices, and ensure compliance with federal mandates and legislation.”[5]
How the IRS Harnesses the Benefits of AI For Enforcement
AI has been useful for the IRS’s enforcement work, including auditing, to reach its strategic priorities. Specifically, as part of its recent strategic operating plan, the IRS leaned into using technology “to be more nimble in the administration of complex tax laws,” and “to evolve along with taxpayer needs and expectations.”[6] The IRS has been using a pilot program to leverage AI for auditing complex partnerships, and recently announced that they opened audits of 76 of the largest partnerships in the U.S. that represent a cross-section of industries.[7]
In addition to audits, the IRS uses AI to combat tax fraud. At the same time, the government and tech experts have warned about the risks that AI poses in generating tax scams[8] so the IRS is working to increase awareness of scams and identity theft, which may be exacerbated by the use of AI. For example, in February 2024, the U.S. Department of the Treasury announced that it recovered over $375 million as a result of implementing, at the beginning of Fiscal Year 2023, an enhanced fraud detection process that utilizes AI.[9]
IRS Guidelines to Mitigate the Risks of Using AI for Enforcement
To mitigate risk in enforcement, it is critical that the IRS develop non-discriminatory auditing practices consistent with its guidelines and additional guidance, such as the White House’s Algorithmic Discrimination Protections.[10] The IRS Interim Guidance requires ongoing monitoring and mitigation for AI-enabled Discrimination, as well as a Risk of Bias Assessment to evaluate the risk of AI operating in a biased manner, “particularly regarding protected groups or classes.”[11]
Research has shown that Black taxpayers are audited by the IRS at higher rates than taxpayers of other races.[12] The IRS Commissioner Danny Werfel confirmed that the IRS “processes were resulting in a disparate impact on Black taxpayers,” and committed to making changes to substantially reduce the number of audits where racial disparities were identified and to the algorithms specifically intended to address the disparity.[13] This is an area where using AI with a specific focus on the measures to identify the impact on equity and the mandate to take action to mitigate discrimination could be transformative. As Werfel has suggested, AI might provide a useful tool for addressing such bias, but should be deployed cautiously, and in a manner mindful of maintaining privacy and security.[14]
The IRS Interim Guidance requires the IRS to assess the impact of AI use, including both its expected benefits as well as the potential risks.[15] The IRS is expected to use risk mitigation practices to mitigate AI-enabled discrimination, such as through “real world performance testing,” independent evaluations and ongoing monitoring and risk assessment.[16] The IRS policies require that humans be trained and have oversight and accountability over decisions or actions.[17] In addition, the IRS is required to identify and assess AI’s impact on equity and fairness and take action to mitigate discrimination.[18]
The U.S. Government Accountability Office (GAO)[19] has stated, “while IRS’s efforts with AI show promise, we think it could do more to document these AI models to ensure they are transparent and implemented consistently in the coming years. Otherwise, IRS risks not being able to explain how it selected taxpayers for these more burdensome audits.”[20] Similarly, this summer, the GAO noted several areas in which the IRS could better document its AI models to more closely align with the GAO’s AI accountability framework and made recommendations regarding documenting key components and technical specifications of the new AI method to select taxpayers to audit.[21] As the IRS moves forward with utilizing new AI and improving on the models, it should do so consistent with its guidelines and goals for responsible AI use and monitoring to mitigate risk.
What Can Nonprofit Entities Expect Moving Forward
As the IRS continues to expand its use of AI for auditing and review, the AI models can be expected to become more efficient and better at detection. Initially, the IRS prioritized high income cases and the largest and most complex partnerships for leveraging its AI enforcement. Moving forward, it will likely apply these AI tools to audit and review an increasing number of entities, including smaller entities such as nonprofits. Nonprofits should continue to ensure that they are complying with reporting and filing requirements to reduce the risk of an audit. As the technology and guidance around AI use continues to rapidly develop, it will be critical for entities to be aware of any changing expectations or requirements. If there are any specific questions about what this means or how to ensure compliance, please contact an attorney.
[1] https://www.whitehouse.gov/briefing-room/presidential-actions/2023/10/30/executive-order-on-the-safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence/
[2] https://www.irs.gov/pub/foia/ig/spder/interim-guidance-raas-10-0524-0001-artificial-intelligence-governance-and-principles-redacted.pdf
[3] https://www.irs.gov/pub/foia/ig/spder/interim-guidance-raas-10-0524-0001-artificial-intelligence-governance-and-principles-redacted.pdf
[4] IRM 10.5.1 Privacy Policy for Artificial Intelligence, available at https://www.irs.gov/pub/foia/ig/spder/pgld-10-0924-0020-public.pdf (last accessed October 8, 2024).
[5] Interim Guidance for New IRM 10.24.1, Artificial Intelligence (AI) Governance and Principles, 10.24.1.1 Program Scope and Objectives, https://www.irs.gov/pub/foia/ig/spder/interim-guidance-raas-10-0524-0001-artificial-intelligence-governance-and-principles-redacted.pdf
[6] IRS 2024 IRA Strategic Operating Plan, Annual Update Supplement, at p. 3, available at https://www.irs.gov/pub/irs-pdf/p3744a.pdf (last accessed October 8, 2024).
[7] IRS releases Strategic Operating Plan update outlining future priorities; transformation momentum accelerating following long list of successes for taxpayers, May 2, 2024; see also IRS announces sweeping effort to restore fairness to tax system with Inflation Reduction Act funding; new compliance efforts focused on increasing scrutiny on high-income, partnerships, corporations and promoters abusing tax rules on the books
[8] https://www.grassley.senate.gov/imo/media/doc/hassan_grassley_to_irs_-_ai_tax_scams.pdf
[9] https://home.treasury.gov/news/press-releases/jy2134
[10] https://www.whitehouse.gov/ostp/ai-bill-of-rights/algorithmic-discrimination-protections-2/
[11] 10.24.1.8.11; 10.24.1.5.2(1)(d) https://www.irs.gov/pub/foia/ig/spder/interim-guidance-raas-10-0524-0001-artificial-intelligence-governance-and-principles-redacted.pdf
[12] https://siepr.stanford.edu/publications/working-paper/measuring-and-mitigating-racial-disparities-tax-audits ; see also https://www.gao.gov/products/gao-24-106126
[13] https://www.irs.gov/newsroom/commissioner-danny-werfel-1-year-anniversary-speech-the-future-irs-transforming-for-a-digital-world
[14] https://www.govexec.com/management/2024/04/irs-commissioner-indicates-ai-will-play-growing-role-future-tax-collection/395867/
[15] https://www.irs.gov/pub/foia/ig/spder/interim-guidance-raas-10-0524-0001-artificial-intelligence-governance-and-principles-redacted.pdf
[16] https://www.irs.gov/pub/foia/ig/spder/interim-guidance-raas-10-0524-0001-artificial-intelligence-governance-and-principles-redacted.pdf
[17] https://www.irs.gov/pub/foia/ig/spder/interim-guidance-raas-10-0524-0001-artificial-intelligence-governance-and-principles-redacted.pdf
[18] https://www.irs.gov/pub/foia/ig/spder/interim-guidance-raas-10-0524-0001-artificial-intelligence-governance-and-principles-redacted.pdf at 10.24.1.8.9 Equity and Fairness
[19] The GAO “provides Congress, the heads of executive agencies, and the public with timely, fact-based, non-partisan information that can be used to improve government and save taxpayers billions of dollars. Our work is done at the request of congressional committees or subcommittees or is statutorily required by public laws or committee reports, per our Congressional Protocols.” https://www.gao.gov/about/what-gao-does
[20] Artificial Intelligence May Help IRS Close the Tax Gap, June 6, 2024.
[21] https://www.gao.gov/products/gao-24-106449 TAX GAP: IRS Should Take Steps to Ensure Continued Improvement In Estimates, Published: May 6, 2024, Publicly Released: June 5, 2024, at pp. 27-31; Artificial Intelligence: An Accountability Framework for Federal Agencies and Other Entities GAO-21-519SP, June 30, 2021; see also Tax Enforcement: IRS Audit Processes Can Be Strengthened to Address a Growing Number of Large, Complex Partnerships GAO-23-106020 Published: Jul 27, 2023. Publicly Released: Jul 27, 2023, https://www.gao.gov/assets/gao-23-106020.pdf