AI in Journalism and Media: Accuracy Obligations, Disclosure, and the Editorial Responsibility Framework

AI is being used across journalism and media — for research, drafting, translation, fact-checking, and content generation. The accuracy obligations, disclosure requirements, and editorial standards that apply.

Key Takeaways

AI-generated journalism that contains false statements of fact creates defamation liability for the publisher — the AI origin of the content does not transfer defamation liability from the publisher to the AI tool provider.
ACMA (Australia), Ofcom (UK), and press standards bodies internationally are developing AI-specific codes for media organisations — the consistent direction is toward disclosure of AI-generated content and human editorial responsibility for all published content.
Copyright in AI-generated content is uncertain — in most jurisdictions, purely AI-generated content without sufficient human creative input may not attract copyright protection. AI-assisted content (where a human author uses AI as a tool) is generally protected, but the line is not always clear.
The hallucination problem is particularly acute in journalism — AI systems that fabricate quotes, invent sources, or confabulate facts create significant defamation and professional reputation risk.
Synthetic media (deepfakes and AI-generated video) that creates false impressions of real people making statements they did not make is regulated in an increasing number of jurisdictions and creates civil liability for creators and distributors.

"情報提供のみを目的としています。この記事は法律、規制、財務または専門的なアドバイスを構成するものではありません。具体的なアドバイスについては、資格を持つ専門家にご相談ください。"

AI in journalism is no longer experimental — it is operational

Newsrooms globally have moved from experimentation to operational AI deployment. AI is now used for: research and source discovery; transcription of interviews and parliamentary proceedings; translation; generation of routine financial reports, sports recaps, and weather summaries; image identification and verification; SEO optimisation; comment moderation; and increasingly, draft writing of news copy. The Reuters Institute Digital News Report 2025 found that over 80% of journalists in major publications use AI at least weekly in their workflow.

This operational shift creates governance obligations that journalism's existing ethical frameworks did not contemplate. The question is no longer whether AI belongs in the newsroom — it does — but how its use is governed to protect the integrity of journalism, the rights of subjects, the trust of audiences, and the legal position of publishers.

The IPSO, IFCN, and editorial framework expectations

Major journalism standards bodies have issued AI-specific guidance. The Independent Press Standards Organisation (IPSO) in the UK applies the Editors' Code of Practice to AI-assisted journalism — Clause 1 (Accuracy) is the most directly engaged, requiring publishers to take care not to publish inaccurate, misleading, or distorted information. If AI generates an inaccuracy and the publisher does not catch it, the publisher is responsible. The International Fact-Checking Network's principles require human verification before publication regardless of whether AI was used in research or drafting.

The major news organisations — Reuters, AP, BBC, Guardian, New York Times, ABC, SBS — have all published AI use policies. The convergent principles across these policies: humans remain editorially responsible for everything published; AI use must be disclosed where it would materially change reader understanding; AI cannot be used to generate quotes or attribute statements to real people; AI cannot generate photorealistic images of identifiable individuals without disclosure; training data sources for any AI tools must be lawfully obtained.

The EU AI Act — transparency obligations for AI-generated content

From 2 August 2026, the EU AI Act's Article 50 transparency obligations require that AI-generated or manipulated content depicting persons, objects, places, or events — which could falsely appear to a reasonable person as authentic — must be disclosed as AI-generated. For journalism, this captures: AI-generated photos or video used as illustration; deepfakes used in reporting (even with disclosure); AI-generated audio in podcasts or radio. AI-generated text published to inform the public on matters of public interest must also be disclosed unless it has undergone human review and editorial responsibility has been assumed.

For publishers operating across EU and non-EU markets, the practical implication is to adopt a single AI disclosure standard that meets EU requirements globally — not to attempt to manage different disclosure regimes per market. Disclosing "AI-assisted" or labelling AI-generated images is now a baseline expectation rather than an editorial choice.

Defamation and AI-generated content

Where AI generates defamatory content — false statements of fact about identifiable persons — the publisher is liable, in every major jurisdiction. The fact that a journalist did not write the statement themselves but used AI to draft it does not shift liability. In Australia, the Defamation Amendment Act 2020 (which took effect across most jurisdictions in 2021) requires a serious harm threshold and contains a public interest defence, but these defences apply to journalism — they do not insulate the publisher from liability merely because AI generated the statement.

AI hallucinations are a particular risk. An AI tool may generate a confidently-stated false claim about an identifiable person (a wrong attribution, a fabricated criminal record, a misattributed statement). If this appears in published journalism, the defamation exposure is real. Mitigations: verify every factual claim independently, especially anything that names individuals; do not rely on AI for source attribution; flag AI-generated material in your editorial workflow so it receives additional fact-checking scrutiny.

Copyright and AI training data — the New York Times v OpenAI question

Journalism organisations are at the centre of the copyright dispute over AI training. The New York Times v OpenAI litigation, initiated December 2023, alleges that OpenAI trained on copyrighted Times content without licence — a case now before US courts with proceedings expected to continue through 2026. Similar litigation is pending against other AI providers from publishers including the Wall Street Journal, New York Daily News, and Le Monde. The UK Copyright, Designs and Patents Act 1988 currently does not provide a text-and-data mining exception for commercial AI training; the EU Copyright Directive provides a TDM exception subject to opt-out.

For journalists, this matters at two levels. First, the AI tools you use may have been trained on copyrighted material — including your own publication's content — without licence. Using those tools does not put you in breach, but the legal status of their outputs (and your publication's outputs created with them) is contested. Second, when you publish AI-generated content, your publication's licence terms with its content management system and syndication partners typically require you to warrant that you own or have licence to the content. AI-generated content may not be copyrightable (US Copyright Office May 2025 report confirmed human authorship requirement), which creates downstream complications.

Practical AI governance for newsrooms

Adopt and publish an AI use policy. The policy should specify: what AI tools are approved for newsroom use; permitted uses (research, transcription, idea generation, headlines) and prohibited uses (generating direct quotes, fabricating images of identifiable persons, drafting investigative reporting conclusions); when AI use must be disclosed in published content; the verification standard before AI-assisted material is published; client/source confidentiality requirements for any AI tool that processes interview transcripts or source documents.

Use enterprise AI tools with training data opt-out. Consumer ChatGPT, free Claude, and free Gemini may retain inputs for training. Enterprise versions (ChatGPT Business, Claude for Work, Microsoft Copilot for Business) have contractual no-training commitments. Sensitive source material — interview transcripts, leaked documents, confidential briefings — should only ever be processed in enterprise-tier tools with no-training commitments confirmed in writing.

Train your journalists. The AI tools are advancing weekly. The journalists who develop AI fluency — understanding the limits, the failure modes, and the appropriate uses — produce materially better work than those who either avoid AI entirely or use it uncritically. Professional bodies (NCTJ in the UK, MEAA in Australia, SPJ in the US) are developing AI training curricula.

Further reading: ISO 42001