AI Taxation: Policy Proposals and IRS Enforcement Shift

The Push for AI Taxation Gains Momentum

Policymakers are increasingly turning their attention to taxing artificial intelligence, driven by concerns over economic inequality and the transformative impact of AI on the workforce. A growing chorus of lawmakers, including Senator Elizabeth Warren, Representative Greg Casar, and Senator Ron Wyden, are floating plans to tax the AI boom. These proposals range from token-based taxes to broader corporate levies, aiming to redistribute the wealth generated by AI and fund social programs.

Senator Warren has been particularly vocal, arguing that overhauling the tax code to tax AI can fund universal healthcare, guaranteed jobs, and accessible education. Representative Casar is calling for a tax on tokens—the units of data processed by AI models—while Senator Wyden is exploring a tax on tech companies to create a wage-security program for workers displaced by automation. These ideas reflect a growing urgency among progressives to address AI's societal impact.

Industry Voices and Political Support

The push for AI taxation is not limited to politicians. Dario Amodei, CEO of Anthropic, has voiced support for robust tax policies on AI, citing the "extreme levels of inequality" he predicts will arise. In a January statement, Amodei argued that such inequality justifies a more aggressive tax approach on moral grounds. This endorsement from a leading AI executive adds weight to the debate, signaling that even within the tech sector, there is recognition of the need for policy intervention.

Interestingly, the concept of "universal basic capital" is gaining traction across the political spectrum. Figures as diverse as Bernie Sanders, Gavin Newsom, Steve Bannon, Sam Altman, and Donald Trump have expressed support for variants of this idea. The policy would give ordinary citizens an ownership stake in firms profiting from AI, reframing the debate from income support to capital distribution. However, implementation challenges remain, including measuring AI-attributable revenue and creating mechanisms for fractionalized ownership.

IRS Codifies AI in Enforcement

On February 10, 2026, the IRS codified its AI enforcement practices into formal policy with IRM 10.24.1. This new Internal Revenue Manual section establishes what AI is authorized to do in the examination process, how mandatory human review is integrated, and what documentation requirements apply to AI-generated referrals. The policy removes any ambiguity about AI being experimental—it is now an operational tool within the IRS.

The IRS now operates two separate AI models to prioritize large partnership returns for examination. For individual returns, AI models select a stratified sample and identify those statistically most likely to contain errors or underreported income. Early pilot results show these new models outperform prior selection methods by a meaningful margin, replacing the older Discriminant Information Function (DIF) system that produced high "no-change" audit rates.

Tax Professionals Face New Liabilities

As AI becomes embedded in both tax preparation and enforcement, tax professionals face a new class of liability. AI systems do not understand tax law—they predict language patterns. This can lead to fabricated court cases, oversimplified interpretations, or incorrect tax positions presented with confidence. The risk is that clients may be led astray by convincing but erroneous AI-generated advice.

Accounting firms are already seeing clients bring in AI-guided strategies. To mitigate risks, professionals are advised to update client intake questionnaires to ask about new entities or strategies implemented without professional consultation. If a client insists on an AI-supported position, firms should document explicit warnings and potential penalties under Section 6662. Establishing clear internal AI policies is also critical, ensuring that any AI-generated research is verified against primary authority.

IRS Codifies AI in Audit Selection

The IRS has taken a significant step by codifying its AI enforcement practices into formal policy. IRM 10.24.1, effective February 10, 2026, establishes what AI is authorized to do in the examination process, how mandatory human review is integrated, and what documentation requirements apply. This removes any ambiguity about AI being experimental—it is now a codified operational tool within the IRS.

The IRS now operates two separate AI models to prioritize large partnership returns for examination. For individual returns, AI models select a stratified sample and identify those statistically most likely to contain errors or underreported income. Early pilot results show these new models outperform prior selection methods by a meaningful margin, replacing the older Discriminant Information Function (DIF) system that produced high "no-change" audit rates.

Tax Professionals Face New Liabilities

AI systems do not understand tax law—they predict language patterns. This can lead to fabricated court cases, oversimplified interpretations, or entirely incorrect tax positions presented with confidence. Tax professionals are already seeing clients bring in AI-guided strategies, creating a new class of liability. Without fundamental knowledge to check AI's work, clients can be led astray by convincing but erroneous outputs.

To mitigate risks, firms should update client organizers to ask about new entities or strategies implemented without professional consultation. If a client insists on an AI-supported position, firms should document explicit warnings and potential penalties under Section 6662. Establishing clear internal AI policies is also essential, ensuring that any AI-generated research is verified against primary authority.

Universal Basic Capital: A New Policy Frontier

The concept of "universal basic capital" is gaining traction across the political spectrum. Advocates argue that giving ordinary people an ownership stake in firms profiting from AI could share the economic upside of automation. This approach reframes the AI-policy debate from income support to capital distribution, potentially changing commercialization incentives for models.

However, implementation requires robust methods to measure AI-attributable revenue and clear legal instruments for fractional claims. Design choices—whether dividends, equity, or tokenization—will determine whether the policy mitigates inequality or creates new concentration and governance risks. For practitioners, these operational details matter more than the slogan, as they interact with model governance and funding flows.

Tax Professionals Navigate New Risks

AI systems do not understand tax law—they predict language patterns. This can lead to fabricated court cases, oversimplified interpretations, or entirely incorrect tax positions presented with confidence. Tax professionals are already seeing clients bring in AI-guided strategies, and the risk of liability is real. Without fundamental knowledge to check AI's work, clients can be led astray by convincing but erroneous outputs.

To mitigate these risks, firms should update client intake questionnaires to ask about new entities or strategies implemented without professional consultation. If a client insists on an AI-supported position, firms should document explicit warnings and potential penalties under Section 6662. Establishing clear internal AI policies is also critical, ensuring that any AI-generated research is verified against primary authority.

IRS AI Enforcement: What Practitioners Need to Know

The IRS's codification of AI in enforcement is a game-changer for tax professionals. IRM 10.24.1 establishes three key things: what AI is authorized to do in the examination process, how mandatory human review is integrated, and what documentation requirements apply. The significance is that AI is no longer experimental—it is a codified policy, with human review structured around AI outputs.

The IRS now operates two separate AI models to prioritize large partnership returns for examination. For individual returns, AI models select a stratified sample and identify those statistically most likely to contain errors. Early pilot results show these new models outperform prior selection methods by a meaningful margin, replacing the older DIF system that produced high "no-change" audit rates. Firms that understand what the IRS's AI is actually doing will be best positioned to provide proactive client guidance.

Operational Risks and Best Practices

Tax professionals face a dual challenge: navigating AI-generated advice from clients and adapting to AI-driven IRS enforcement. The key is to treat AI as an efficiency tool, not an authority. Tax law is highly fact-specific, jurisdiction-dependent, and constantly evolving, relying on nuance that algorithms cannot replicate. Firms should design around the disclosure question, potentially keeping AI processing inside the firm to avoid data transmission issues.

Running AI on hardware the firm controls means taxpayer information never leaves, simplifying compliance with Section 7216. The assumption that sending data to a large outside model is the only way to use AI is becoming outdated. Firms that recognize this now will have cleaner answers sooner, positioning themselves for deeper, year-round advisory relationships that no algorithm can replicate.

Universal Basic Capital: Implementation Challenges

For practitioners, the operational details of universal basic capital matter more than the slogan. Distributing capital claims tied to AI output requires reliable measurement of model-attributable revenue, new accounting conventions, and mechanisms for fractionalized ownership. These mechanisms interact with model governance: if payouts depend on commercial success, firms and investors may accelerate deployment and reduce conservatism around safety controls.

Changes to investor returns or taxation could affect funding flows into compute, data acquisition, and R&D. The design choices—whether dividends, equity, or tokenization—will determine whether the policy mitigates inequality or creates new concentration and governance risks. For now, the debate is shifting from whether to tax AI to how to implement such taxes effectively.