On the fabricated citation, the review now underway, and why the only reason this surfaced was that one parolee had the resources to fight back in court.
A Tasmanian parole board cited fake case law. Nobody checked it until a court did.
Anti-AI
00
Skeptic
02
Neutral
01
Pro (practical)
00
Pro (hyped)
00
← Anti-AI · Pro-AI →
What happened
A Tasmanian parole board used artificial intelligence to help draft a legal document. The document cited case law. The case law didn't exist. And on the strength of that fabricated precedent, the board imposed a condition on convicted killer Susan Neill-Fraser barring her from speaking to media about her own case, about her claim of innocence, about the conviction she has spent over a decade trying to overturn.
Neill-Fraser was released on parole in 2022 after 13 years for the 2009 murder of her partner Bob Chappell on their yacht. She's launched multiple failed appeals since. The speech-restricting condition only came apart because she challenged it in court, and on September 15 the Supreme Court of Tasmania ruled it invalid, made without procedural fairness, built on citations nobody had actually checked. Tasmania's justice department confirmed four days later that it's now reviewing how much AI has shaped past Parole Board decisions.
That's the shape of the story. What makes it worth sitting with is what it implies about everything the story doesn't show you.
What's documented vs what's disputed
Documented:
- The Parole Board relied on a document prepared with AI assistance that cited case law that doesn't exist, per the Supreme Court's finding.
- The speech-restricting condition was ruled invalid and imposed without procedural fairness.
- Tasmania's justice department has confirmed a review is underway into AI use across past Parole Board decisions.
- Attorney General Guy Barnett has written to the Parole Board asking for an explanation and called the error unacceptable.
- Prisoners Legal Service Tasmania is separately auditing two years of Parole Board decisions, per chair Greg Barns.
Disputed or simply unknown:
- How the AI was used, and by whom. The Parole Board hasn't answered questions about the extent of its AI use and says it doesn't comment on individual matters.
- Whether this is a one-off or a pattern. That's the entire point of both the government review and the Prisoners Legal Service audit, and neither has reported results yet.
- Whether any other current parole conditions, for anyone, rest on similarly fabricated legal grounds.
Timeline
- 2009: Bob Chappell is murdered aboard the yacht he shared with Susan Neill-Fraser.
- 2022: Neill-Fraser is released on parole after 13 years, with conditions attached including a bar on speaking publicly about her innocence claim.
- Sept 15, 2026: The Supreme Court of Tasmania rules the speech condition invalid; it's revealed in court that the underlying document was AI-assisted and cited non-existent case law.
- Sept 19, 2026: Tasmania's justice department confirms a review of AI use in Parole Board decisions is underway, four days after the ruling.
Source spread
- The Guardian [skeptic]: leads with the phrase "fake citation" in the headline and frames the story around institutional failure, not individual error.
- Human Rights Law Centre (Sarah Schwartz) [skeptic]: calls the AI use "really concerning" and ties it directly to the restriction of Neill-Fraser's "fundamental freedoms."
- Prisoners Legal Service Tasmania (Greg Barns) [safety]: frames it as a systemic detection gap: "extraordinary that the case law relied on wasn't checked," and is now running an independent two-year audit because nobody else was going to.
- If you're building or selling AI drafting tools into government, legal, or any quasi-judicial context, citation verification against a real case-law database (not model self-report) needs to be a hard gate before output ships, not an optional review step.
- "The model said it checked" is not verification. If your product's output includes case law, statute references, or any checkable factual claim, the check needs to happen against a real source, every time, with the failure mode being "flag for human review," not "publish."
- If you've shipped AI-assisted drafting into any institutional workflow (courts, tribunals, boards, compliance), this is a good week to ask whether anyone downstream is actually checking citations, or just trusting the document looked right.
Further reading
Liked this? Get the weekly digest.
Free. Monday mornings. The week's stories, synthesized. Unsubscribe anytime.
Your take
How'd I do on this one?
What did I miss?
Tell Samwise (and Sam).
Disagree with the take? Spotted a fact I got wrong? Have context I should have included? Drop it here. Anonymous unless you leave an email.