Tantaros v. Fox News: SDNY Warns Former Fox Anchor for AI-Generated Statute Language and Non-Existent Cases

Background: A High-Profile Employment Dispute

Andrea Tantaros, a former Fox News anchor, filed suit pro se against Fox News Network and several individual defendants including Roger Ailes. The case raises gender-based violence and employment law claims. In opposing defendant Scott Brown’s motion to dismiss, Tantaros filed an opposition brief containing both inaccurate and non-existent citations.

When Brown’s counsel notified Tantaros of the citation issues, she filed a “Notice of Correction of Citations” that voluntarily withdrew some — but not all — of the problematic citations. She then filed an unauthorized sur-reply that contained additional inaccurate citations. This post-correction recurrence is the pattern that triggered the court’s inference of AI use.

The AI Issue: Fabricated Statute Text + Non-Existent Cases

The AI hallucinations in Tantaros’s filings operated at two levels. First, she quoted non-existent language from the Gender-Motivated Violence Protection Law (VGMVPL/GMVA), attributing to the statute text that does not appear in it: “[t]he GMVA expressly applies to ‘assault sexual misconduct, sexual abuse, and other forms of gender-based violence, including emotional abuse and harassment.'” The statute does not contain this language. Second, she cited multiple non-existent cases including “M.D. v. New York State Off For People with Developmental Disabilities, 2023 WL 3404357 (S.D.N.Y. 2023)” and others.

The court inferred AI tool use: “This pattern of behavior indicates that Tantaros has used artificial intelligence tools in preparing her filings but did not verify the accuracy of citations produced by those artificial intelligence tools before submitting them to the Court.”

Holdings

1. Sur-reply struck. The unauthorized sur-reply was struck on procedural grounds (unauthorized) and separately noted for its inaccurate citations.
2. Partial correction insufficient. Tantaros’s voluntary withdrawal of some (but not all) non-existent citations did not prevent the court from noting the AI issue and issuing a warning.
3. AI use inferred from pattern. The court explicitly stated that the pattern — inaccurate citations in two filings, including after partial correction — indicates AI use without proper verification.
4. Future violations may result in sanctions. The court issued an express warning that any future filings containing inaccurate or non-existent citations may result in sanctions.

“This pattern of behavior indicates that Tantaros has used artificial intelligence tools in preparing her filings but did not verify the accuracy of citations produced by those artificial intelligence tools before submitting them to the Court. Like represented litigants, pro se litigants such as Tantaros have a duty not to submit false citations to the Court.”

— Judge Sidney H. Stein, S.D.N.Y., March 27, 2026

India Angle: Fabricating Statute Text — A Distinctive AI Risk

Tantaros’s case illustrates an AI hallucination variant that is particularly dangerous in Indian legal practice: fabricating statutory language. India’s complex multi-level statute corpus — Central Acts, State Acts, Rules, Regulations, and subsidiary legislation — is a rich source of AI confusion. An AI tool that generates plausible-sounding language attributed to a real statute provision causes immediate harm because courts typically accept quoted statutory text without cross-checking the bare Act.

Relevant Indian Law

• Any statute citation in Indian proceedings: The standard practice in Indian courts is to read statute provisions from a certified copy or officially published version (Gazette of India). If an advocate quotes a provision that does not appear in the Act, this is a misrepresentation of law — engaging BCI Rule 9 and potentially contempt jurisdiction.
• POSH Act 2013 / Domestic Violence Act 2005 parallel: Tantaros’s case involved gender-based violence law. Indian POSH Act proceedings before Internal Complaints Committees and Local Committees frequently involve self-represented complainants. AI-generated POSH Act quotations that do not appear in the Act could undermine otherwise valid complaints.

Three Practical Tips

1. Verify every statute quotation against the bare Act text. Always open the bare Act on India Code (indiacode.nic.in) and confirm the quoted language appears verbatim before including it in a submission. AI-generated statute quotations often blend language from the Act’s preamble, objects, definitions, and substantive provisions into plausible-sounding but composite text.
2. A partial correction that leaves some errors in place is not a sufficient remedy. The Tantaros pattern — voluntarily withdrawing some citations but not all — is exactly the kind of half-measure that courts treat as inadequate. In India, once a citation error is identified by opposing counsel or the court, the corrective duty extends to a complete review of all citations in that and subsequent filings.
3. High-profile cases attract closer judicial scrutiny of citation accuracy. Tantaros vs. Fox News is a high-profile proceeding. Indian high-profile cases — RTI matters, PILs in the Supreme Court, celebrity divorce proceedings — similarly attract closer judicial attention. AI errors in high-profile Indian proceedings are more likely to be noticed and more likely to be published.

Quick Takeaways

• AI hallucinations can fabricate statutory language, not just case citations — this is a distinct and dangerous variant because courts accept quoted statutory text more readily than they verify case citations.
• A partial voluntary correction that leaves some errors in place does not prevent a judicial AI inference or a sanctions warning.
• The pattern of errors across two filings (including one filed after partial correction) is sufficient for courts to conclude AI tools were used without proper verification.
• High-profile pro se litigants in federal court are held to the same verification standard as represented parties.
• In India: fabricating statutory text in a proceeding is among the most serious forms of AI misconduct, potentially engaging contempt and BCI disciplinary jurisdiction simultaneously.