⚡ CASE DIGEST
Emmanuel S. Yirenkyi v. Angela Hoover, Warden — M.D. Pennsylvania, 2 February 2026
A Pennsylvania federal court ordered a bond hearing for Emmanuel Yirenkyi — a Ghanaian permanent resident detained pending removal — while noting LLM hallucinations in the habeas brief. The court addressed the AI-generated non-existent citations while granting relief on the underlying immigration detention claim.
Why it matters: LLM hallucinations now appear in immigration habeas petitions filed on behalf of detained non-citizens. The stakes could not be higher — and yet AI-generated fabricated citations are being filed in cases involving people’s liberty.
Category: Hallucinations & Sanctions | Jurisdiction: USA | Read time: 5 min
Case at a Glance
| Full Citation | Emmanuel S. Yirenkyi v. Angela Hoover, Warden, No. 3:25cv2414 (M.D. Pa.) |
| Court | United States District Court, Middle District of Pennsylvania |
| Date of Decision | 2 February 2026 |
| Category | AI Hallucinations & Sanctions |
| Jurisdiction | USA |
| AI Tool Used | Large Language Model / LLM (identified by court) |
| Judgment / Order | 2026 WL 268230 |
Background
Emmanuel Yirenkyi, a Ghanaian citizen and US permanent resident, was detained by immigration authorities after pleading guilty to fraud offences and facing removal. His habeas petition challenged his detention. The brief cited cases including references to ‘German Santos v. Warden Pike County Correctional Facility’ in support of due process arguments about bond hearings. While that specific case exists, other cited authorities in the petition appeared to be LLM hallucinations.
The AI Issue
The court identified large language model hallucinations in the habeas brief — cases cited that did not exist in any legal database. The court used the specific term ‘LLM’ in its published decision, reflecting growing judicial familiarity with the technology. Despite the citation issues, the court found that due process compelled a bond hearing and ordered one — but the AI citation problem was documented on the record.
- The court ordered a bond hearing with an Immigration Judge in accordance with German Santos v. Warden — relief granted on the merits [habeas relief granted].
- The court identified LLM hallucinations in the petition’s briefing — specific citations to non-existent authority [LLM terminology used].
- The use of ‘LLM’ in a published immigration habeas decision is significant — courts are demonstrating specific technological knowledge [judicial AI literacy].
- Despite the AI citation errors, the court proceeded to grant the substantive relief sought [fabrications did not doom the petition].
- The case confirms that immigration habeas practitioners are among the highest-risk groups for AI citation misconduct [detention bar warning].
“The Court addressed the hallucinated citations identified in petitioner’s brief, including references to large language model-generated authority that did not exist in the legal database.”
— M.D. Pennsylvania, Yirenkyi v. Hoover, 2 February 2026
The India Angle
Indian Law Equivalent: Under Rule 15 of the Bar Council of India Rules and the Contempt of Courts Act, 1971, submitting fabricated or non-existent case citations before any Indian court constitutes professional misconduct and contempt. The duty to verify cited authorities is absolute — AI does not create an exception.
Bar Council Rules: BCI Rule 15 (duty not to mislead the court) and Rule 33 (duty to conduct cases with integrity) are directly engaged when AI-generated hallucinated citations are filed. Disciplinary proceedings before the State Bar Council can result in suspension or removal from the roll.
Practical Advice for Indian Advocates: Verify every AI-generated citation on SCC Online, Manupatra, or IndianKanoon before filing. No AI tool — ChatGPT, Claude, Gemini, or any Indian legal AI — reliably produces accurate citations. Treat AI output as a starting point for India research, not a final product.
Quick Takeaways
- LLM hallucinations in immigration detention habeas petitions place a person’s liberty at risk — the highest possible stakes for AI citation errors.
- Courts are using the term ‘LLM’ in published decisions — judicial AI literacy is now a measurable fact, not an assumption.
- Indian advocates handling habeas corpus petitions under Article 32 and 226 must apply the strictest possible verification standard to AI-generated citations.