Beato-Estrella v. Swaney: AI Hallucination Warning Issued to Incarcerated Pro Se Petitioner

Background: An Incarcerated Petitioner, BOP Programming, and an ICE Detainer

Beato-Estrella, an incarcerated individual at a federal facility in the Middle District of Pennsylvania, filed a pro se petition challenging certain aspects of his confinement — specifically, issues relating to Bureau of Prisons programming access and an ICE detainer that he alleged was depriving him of eligibility for certain programs. The court found it lacked subject matter jurisdiction: Beato-Estrella had not shown that any aspect of his original judgment was being ignored by the BOP, and he was not currently eligible for any program from which the ICE detainer was depriving him access. The petition was dismissed without prejudice.

Before dismissing, the court addressed the quality of Beato-Estrella’s legal research. He had cited cases that were entirely irrelevant to his petition — relating to the conviction itself rather than the execution of the sentence. The court used this as an opportunity to issue a general AI warning.

The AI Issue: Hallucination Warning to an Incarcerated Pro Se Petitioner

Judge Neary’s AI warning is particularly thoughtful in its acknowledgment of the specific barriers facing incarcerated litigants: “The court understands legal research is difficult for anyone, especially when considering the constraints of doing so while incarcerated.” This is not an academic point — access to legal materials in federal prisons is highly limited. Incarcerated individuals often have access only to legal aid software (TRULINCS), outdated physical law libraries, or, increasingly, devices with AI tool access.

The judge specifically noted: “AI programs have been known to hallucinate, or create fake cases, in response to a user’s query,” citing Benjamin v. Costco Wholesale Corp., 779 F. Supp. 3d 341 (E.D.N.Y. 2025). Despite the empathetic framing, the court did not relax the standard: “even though the court affords latitude to parties who are pro se, they must still ensure to follow the rules of the court and cite to actual law.”

Holdings

1. Petition dismissed without prejudice. No subject matter jurisdiction existed on the facts presented.
2. AI hallucination warning issued with empathy. The court acknowledged the difficulty of legal research while incarcerated before issuing the warning — a more empathetic framing than most AI hallucination orders.
3. Standard unchanged. Pro se status, including for incarcerated individuals, does not exempt a litigant from the requirement to cite actual law.

“AI programs have been known to hallucinate, or create fake cases, in response to a user’s query… The court understands legal research is difficult for anyone, especially when considering the constraints of doing so while incarcerated. Still, even though the court affords latitude to parties who are pro se, they must still ensure to follow the rules of the court and cite to actual law.”

— Judge Keli M. Neary, M.D. Pennsylvania, March 16, 2026

India Angle: AI and Legal Research for Incarcerated Persons in India

India’s prisons house over 550,000 inmates (NCRB 2022), many of whom file bail applications, habeas corpus petitions, and appeals pro se or with minimal legal assistance. Legal Aid Services under Section 12 of the Legal Services Authorities Act 1987 are available but implementation is uneven. As mobile devices and tablet programs enter some Indian prisons, AI tool access is an emerging reality.

Relevant Indian Law

• Article 21, Constitution of India / CrPC Section 436A: The right to legal assistance for incarcerated persons is a fundamental right. If AI tools are used to exercise that right, the same verification standards apply as for represented parties.
• National Legal Services Authority (NALSA): Legal aid advocates representing incarcerated persons must not rely on AI-generated citations without verification. The duty to the client requires accurate legal research, not just arguable submissions.

Three Practical Tips

1. Legal aid advocates must be especially rigorous with AI verification for prison clients. An incarcerated client cannot independently verify citations; the advocate must do so. AI errors in habeas or bail applications can lead to dismissal, denial, or, worse, discourage courts from treating genuine claims seriously.
2. Relevance checking is the first line of defence against AI hallucination. The Beato-Estrella error was not just fabricated citations but irrelevant citations — cases about conviction rather than sentence execution. A basic relevance check (does this case actually address my issue?) will catch many AI hallucinations before they reach the court.
3. Teach clients in legal aid clinics about AI limitations before they use public-access tools. As digital courts and e-filing expand in India, incarcerated or unrepresented persons may increasingly use public-access AI tools. Legal aid clinics should include AI verification training in their client education programmes.

Quick Takeaways

• AI hallucination warnings are now reaching incarcerated pro se litigants — one of the most resource-constrained groups of court users.
• Judicial empathy for research constraints does not change the verification standard — even incarcerated litigants must cite actual law.
• Citing entirely irrelevant cases (right topic, wrong legal category) is an AI hallucination pattern distinct from citing nonexistent cases.
• In India: legal aid advocates and NALSA representatives bear the verification duty for incarcerated clients who cannot check citations themselves.