AI Hallucinations Kept a Convicted Woman Silent, Case File Shows
Tasmania's Parole Board cited nonexistent legal authorities to stop Susan Neill-Fraser from publicly maintaining her innocence in a 2009 murder case.

Updated
Why it matters
- Tasmania's Parole Board used a document citing legal authorities that do not exist to prevent Susan Neill-Fraser from proclaiming her innocence
- Bob Chappell, a 65-year-old physicist at the Royal Hobart Hospital, disappeared from his 53ft ketch the Four Winds on the Derwent River on 26 January 2009
- Chappell had bought the yacht for $200,000 less than four months before his disappearance; Neill-Fraser was convicted of his murder and says she is innocent
- Advocates describe the case as 'deeply troubling'
Tasmania's Parole Board relied on a document citing legal authorities that do not exist to stop Susan Neill-Fraser from publicly maintaining her innocence — and advocates say the fabricated citations bear the hallmarks of AI hallucination.
The case has become the latest flashpoint in a debate that courts, parole boards, and tribunals around the world are now confronting: what happens when generative AI tools, which are known to invent case law, citations, and legal authorities, seep into consequential decisions about a person's liberty and speech.
The disappearance
The case begins on 26 January 2009. A yacht called the Four Winds sat moored on the Derwent River in Tasmania. The vessel was, by all accounts, a troublesome one.
Bob Chappell, a physicist at the Royal Hobart Hospital, was on board the 53ft ketch that day. The 65-year-old was trying to fix yet another issue with the yacht — a boat he had bought for $200,000 less than four months earlier.
Chappell disappeared from the Four Winds. Susan Neill-Fraser was subsequently convicted of his murder. She has maintained for years that she did not commit the crime.
The hallucinated document
Neill-Fraser's conviction is only part of the story. The recent development concerns what happened after: the Tasmanian Parole Board used a document that cited legal authorities which do not exist to prevent her from speaking out about her innocence.
Legal authorities that cannot be found in any law report are a signature failure mode of large language models. When generative AI systems produce legal text, they routinely invent plausible-sounding case names, citations, and precedents that have no basis in the actual record. Lawyers in multiple jurisdictions have been sanctioned for filing court documents containing such fabricated citations.
In this instance, the fabricated authorities appear in paperwork used to justify restricting the speech of a person convicted of murder — a person who says the conviction itself is wrongful.
Advocates following the case have described the situation as "deeply troubling."
Why this matters
The Neill-Fraser case sits at the intersection of two high-stakes problems. The first is the integrity of post-conviction processes. Parole decisions determine whether a person regains freedom, under what conditions, and with what restrictions on their conduct. If the paperwork informing those decisions contains invented legal authorities, the entire process loses its factual foundation.
The second is the free speech dimension. Neill-Fraser wanted to proclaim her innocence publicly. A document citing nonexistent law was used to stop her. For wrongful-conviction advocates, that sequencing is alarming: the tool that was supposed to ground the Board's decision instead introduced fabricated law into a proceeding that constrained a person's ability to argue she should never have been convicted at all.
The case also lands at a moment when Australia's legal institutions, like their counterparts elsewhere, are scrambling to establish rules for AI use. Courts in several jurisdictions now require lawyers to certify that AI-generated material has been verified. Parole boards and administrative tribunals, which operate with less scrutiny than open courts, have largely escaped that reckoning — a gap the Neill-Fraser case exposes.
The pattern
AI hallucinations in legal settings are not hypothetical. Since generative AI tools became widely available, cases have surfaced of filed briefs, judgments, and internal documents containing invented citations. Each incident has raised the same question: who checked the work, and what happens to the person on the other end of the fabricated document when nobody did?
The Tasmania case sharpens that question because the person on the other end was a prisoner with no independent platform, dependent on the fairness of the very body that used the flawed document against her.
The full details of the case — including the exact origin of the document, how it came before the Parole Board, and what the Board has said in response — are laid out in the Guardian's continuing coverage of the matter.
What comes next
Advocates say the case demands answers from the Tasmanian Parole Board about how a document citing nonexistent legal authorities entered its process, and what safeguards exist to prevent a recurrence. For Susan Neill-Fraser, the immediate stakes are concrete: the right to say, in public, that she did not kill Bob Chappell. For the wider legal system, the case is a test of whether institutions that make decisions about liberty — away from the scrutiny of open court — are equipped to catch the confident fabrications that generative AI produces.
Source: The Guardian AI
More from Rebecca Stone
Show full bio
Correspondent covering consumer brands and retail at AI In Context.
135 articles
Related articles
- Florida AG asks court to halt OpenAI model training without oversight
- AI 'Godfathers' Hinton and Bengio Warn Governments of Intelligence Explosion
- Jensen Huang Calls AI Distillation 'Competition' as White House Calls It 'Theft'
- DHS Will Use Google AI to Redact FOIA Documents
- Wuhan Court Counts AI Token Costs in Copyright Damages