Fake papers, real names
On 8 July 2026, Newsroom reported that a 71-page Federated Farmers report on controlling feral deer populations contained fabricated academic citations produced through AI hallucinations. Policy adviser Phil Holland confirmed he used AI to compile the references and accepted responsibility for the errors. The damage runs in two directions. Scientists found their names attached to research they never published, a straightforward reputational harm. And a formal policy document from one of the country’s most influential rural lobby groups now rests, in part, on citations that do not exist.
This is not a story about a bad actor. Holland acknowledged what happened and appears to have acted in good faith. The problem is structural, and it is bigger than one report.
Why the pipeline matters
Federated Farmers policy papers are not academic curiosities that sit on a shelf. They are read by Ministry for Primary Industries officials, regional councils, and parliamentary select committees. Feral deer control policy feeds directly into biosecurity levies on farmers and landowners, permitted pest control methods and chemical approvals through the Environmental Protection Authority, land use restrictions, and regional council pest management plans that ratepayers fund.
That is the chain that matters for business. If a recommendation rests on a citation that was never written, the regulation that follows is built on fiction, but the compliance cost is entirely real. Farmers pay it. Processors adjust to it. Councils enforce it. And the person who wrote the report has moved on to the next brief.
Why this keeps happening
The hallucination problem is not a glitch waiting for a patch. Large language models are built to produce fluent, confident text, and when asked for a reference list they will happily generate plausible-looking author names, journal titles, years and volume numbers that correspond to nothing. As Oxford Internet Institute professor Sandra Wachter warned in 2024, these models are “trained to be engaging and profitable” but “whether they speak the truth or not is not of concern.” Wachter and colleagues have argued for a new legal framework to hold model developers accountable for what they call “careless speech.”
The legal profession learned this the expensive way, when US lawyers filed briefs citing court cases that did not exist. Agricultural policy is now learning the same lesson, only without a judge to throw the document out.
Slop with a letterhead
The Federated Farmers case does not sit in isolation. It lands inside a documented pattern of AI-generated content taking root in New Zealand’s information environment. In early 2026, AUT associate professor Merja Myllylahti published a baseline report finding AI tools are already widely used in NZ newsrooms, particularly in commercial media, and that 60 per cent of New Zealanders are uncomfortable with news produced mostly by AI under limited human oversight.
RNZ and TVNZ have investigated fake news sites such as NZ News Hub, which carried no original reporting and used unlabelled AI-generated images, including doctored pictures of the Akaroa Harbour boat incident. As journalist Nik Dirga wrote in February 2026, “AI slop is everywhere on social media now, literally any time a news event happens, a horde of pages serve up AI-generated garbage about it instantly.”
The difference here is the letterhead. A Facebook page carries no authority. A Federated Farmers submission does. When slop wears an institutional badge, officials read it as evidence.
The verification gap nobody owns
Myllylahti’s report also found some NZ editors regard tagging AI content as unnecessary, on the logic that ChatGPT is really just a replacement for a Google search. Apply that thinking to policy documents, and you get exactly what the Federated Farmers case exposed. No disclosure, no verification requirement, no audit trail.
Here is the uncomfortable part. No government agency currently requires submitters to certify that citations in a policy document are real. There is no MPI or EPA checklist item asking whether every cited paper actually exists. Peer review, editorial oversight and formal verification, the three filters that would normally catch this, are all absent from the policy submission pipeline.
That gap now has a name attached to it. The question is who closes it. Federated Farmers can impose its own verification standard on anything carrying its name, and it should. MPI and the EPA can require submitters to attest that sources are genuine. Parliament can demand the same of select committee submissions. Until one of them does, businesses are exposed to the possibility of paying for rules that were partly written by a machine that cannot tell truth from a convincing sentence.
Sources
- Newsroom: Pest control report cites fake papers, relying on ChatGPT hallucinations (2026-07-08)
- Knightly Views: Another threat to our news industry, AI slop (2026-02-10)
- Nik Dirga: Breaking, the never-ending battle against fake AI news (2026-02-15)
- Reuters Institute: AI-generated slop is quietly conquering the internet (2024-11-26)