AI assists. The record decides.

AI is part of NULLPEN’s production system, not a substitute for evidence. It can increase the breadth, speed and consistency of review. It can also introduce confident errors, source drift, bias and false precision. The controls must evolve as the tools and risks change.

• Inventing, reconstructing or disguising a source or quotation.
• Publishing an accusation solely because a model surfaced it.
• Fabricating a photograph, voice recording or eyewitness scene as evidence of a real event. A disclosure does not make fabricated evidence acceptable.
• Generating a person’s voice or likeness without appropriate rights and context.
• Turning sensitive personal data into behavioral, safety or investment conclusions without a lawful, reviewed basis.
• Automatically publishing a correction, retraction or high-impact forecast without human approval.
• Hiding material model uncertainty behind polished language or a numerical confidence score.

Source review

A reviewer must be able to inspect the records behind material claims and distinguish source text from model synthesis.

Editorial approval

Publication state, headline, regional implication and forecast label remain accountable editorial decisions.

Conflict review

Material discrepancies between sources and generated summaries are resolved against the record, not by majority model vote.

Audit trail

High-impact workflows should preserve model/provider version, date, input provenance, reviewer and substantive changes where feasible.

Security

Sensitive or embargoed material is not placed into an external AI service without an approved data-handling arrangement.

The separately labeled Device voice reader is an accessibility option using an installed English speech voice. It is not NULLPEN Reader 01 or 02, does not impersonate a house narrator and does not create a reusable podcast or social-video recording. Playback is user initiated; report eligibility and correction checks remain in force.

NULLPEN may offer separately licensed synthetic narrators for article audio and short news recaps during the founding phase. Each voice is a disclosed production interface, not a fictional reporter or eyewitness. A reader may choose a narrator; NULLPEN does not infer that preference or silently substitute another voice. Scripts inherit the evidence state and correction obligations of the underlying record.

Every AI-generated illustration must be labeled at the image, including social and recap versions. Materially altered visuals must explain changes that affect their interpretation. Concept art is not documentary capture. Sound effects, motion graphics and presenter treatments may add clarity and energy; they may not manufacture certainty, urgency or testimony.

Each voice needs its own documented permission for the actual provider, channels and uses. Editorial permission does not automatically cover advertising or political messages. Provider terms, input rights and output review must be recorded; an AI label neither guarantees non-infringement nor establishes exclusive copyright.

English is the editorial master for the initial platform. Spanish subtitles and text translations should preserve claim state, uncertainty, attribution and correction notices. Where wording could change legal, safety or financial meaning, human bilingual review is required before publication.