**The U

The U.S. Tax Court handles over 30,000 cases annually, and a disproportionate number involve self-represented taxpayers who are already turning to AI for help. That creates a two-front problem: attorneys need guardrails for AI-assisted tax filings, and the court needs a framework for the flood of pro se submissions generated by ChatGPT. Bloomberg Law reported in 2025 that the Tax Court is actively navigating AI misuse rules specifically because of the self-represented filer problem.

Blue J raised $122 million in Series D funding to build what it calls an 'AI operating system for tax cognition' — and that tells you where the market is heading. The platform uses RAG architecture trained on federal court decisions, IRS rulings, and Treasury regulations to predict tax outcomes with claimed accuracy rates exceeding 90%. But the 2026 Heppner ruling just held that communications with publicly available AI platforms aren't protected by attorney-client privilege — meaning every AI-assisted tax memo your team generates could be discoverable.

Tax Court AI Rules: What Exists Today

The U.S. Tax Court hasn't issued a blanket AI disclosure requirement, but it's moving in that direction. Bloomberg Law reported that the court is developing specific guidance for AI misuse, driven by two converging pressures: the high volume of pro se filers who use AI without understanding its limitations, and the technical complexity of tax filings where AI errors are harder to detect. The court already requires that all submissions comply with Rule 33 (signatures and representations) and Tax Court Rule 123 (sanctions for frivolous arguments). Judges are now applying these existing rules with AI-specific scrutiny. If an attorney submits a brief citing a Tax Court memorandum opinion that doesn't exist, Rule 123 sanctions apply regardless of whether a human or an AI fabricated the citation.

Blue J and the AI Tax Intelligence Arms Race

Blue J's $122 million Series D — co-led by Sapphire Ventures — positions it as the dominant AI platform for tax research and prediction. The platform is built on a Retrieval-Augmented Generation system using GPT-4.1, trained exclusively on publicly accessible, authoritative sources: federal court decisions, IRS rulings, Treasury regulations, and expert commentary from Tax Notes. Blue J claims its predictive models can forecast tax outcomes with over 90% accuracy for well-defined fact patterns. That's compelling for managing partners evaluating whether to license the platform. But the accuracy claim comes with a critical caveat: it works best for pattern-matching against established precedent. Novel tax positions, aggressive planning structures, and emerging regulatory areas — exactly the situations where firms earn premium fees — remain firmly in the domain of human judgment. Use Blue J for research efficiency, not as a substitute for tax expertise.

The Heppner Privilege Problem

The 2026 Heppner ruling is a wake-up call for every tax practice using AI. The court held that a defendant's communications with a publicly available AI platform were protected by neither attorney-client privilege nor work product doctrine. For tax practitioners, this means every query you type into ChatGPT, Claude, or any non-privileged AI platform about a client's tax position is potentially discoverable in subsequent litigation or IRS examination. The practical implication is stark: if you're using general-purpose AI to brainstorm tax strategies, analyze exposure, or draft position papers, those interactions could be produced in discovery. The fix isn't to stop using AI — it's to use privileged, enterprise AI platforms that operate under your firm's confidentiality framework, with BAAs and data processing agreements that maintain privilege protection.

Pro Se Filers and AI: The Court's Biggest Headache

Self-represented taxpayers file roughly 60% of Tax Court petitions, and an increasing number are using AI chatbots to draft those filings. The results are predictable: petitions citing nonexistent Tax Court opinions, arguments based on debunked 'tax protestor' theories repackaged in professional-sounding language, and computational errors presented with false precision. The Tax Court faces a unique challenge here that other federal courts don't. Unlike district courts where pro se civil filings are a fraction of the docket, the Tax Court's pro se volume is the majority of its caseload. Any AI policy the court adopts will disproportionately affect unrepresented taxpayers — many of whom are using AI precisely because they can't afford an attorney. The court is walking a tightrope between access to justice and filing quality.

Compliance Framework for Tax Practices Using AI

Managing partners running tax practices need a three-part AI compliance framework. First, platform selection matters. Use enterprise AI tools that sign BAAs, don't train on your data, and maintain privilege protection — not consumer-grade chatbots. Blue J, Harvey, and CoCounsel all offer tax-specific functionality with enterprise security. Second, build verification workflows for tax-specific risks. AI-generated tax research must be checked against current Treasury regulations, IRS guidance, and Tax Court precedent. Tax law changes faster than most AI training data — the TCJA, SECURE Act, Inflation Reduction Act, and annual revenue procedures create a constantly shifting landscape. Third, document your AI usage for each engagement. The Heppner ruling means your use of AI tools could become a discovery issue. Maintain logs showing which tools were used, what inputs were provided, and what human review was performed. That documentation is your defense against both malpractice claims and court sanctions.

The Bottom Line: The Tax Court is developing AI-specific guardrails driven by a unique combination of technical complexity and high pro se volume. Blue J is the market leader for AI-powered tax research, but the Heppner ruling means privilege protection for AI communications is not guaranteed. Tax practices should use enterprise platforms with BAAs, verify every AI-generated citation against current authority, and document their AI workflows as if every interaction will be reviewed in discovery — because after Heppner, it might be.