What you need to know
AI image generators — including Adobe Firefly, Midjourney, and DALL-E — can produce photorealistic visuals indistinguishable from documentary photography. For UK journalists, using such images without clear disclosure risks a breach of IPSO Editors' Code Clause 1 (Accuracy), which prohibits misleading or distorted content. The NUJ's published AI guidance calls on members to disclose “significant AI use” in any content they produce, including visual material.
The Coalition for Content Provenance and Authenticity (C2PA)has developed the Content Credentials standard — tamper-evident metadata embedded in image files that records whether and how AI tools were used in creating or editing the image. Leading platforms are beginning to embed these credentials by default, and the Content Credentials Verify tool allows anyone to inspect an image's provenance chain before publication.
The Reuters Trust Principles require transparency about how content is produced. UK newsrooms are expected to develop internal AI imagery policies that cover: labelling conventions visible to readers, verification workflows before publishing AI-sourced or AI-altered visuals, and escalation procedures when deepfake or manipulated imagery is suspected.
When AI-generated imagery disclosure issues arise
- 1Using an AI-generated image to illustrate a news story about a real event without labelling it as AI-generated — creating a false impression of documentary evidence.
- 2Receiving an image from a source or wire feed that has been deepfake-altered to misrepresent a public figure's appearance, location, or actions.
- 3A picture desk sourcing AI-generated stock imagery through a library that does not flag AI origin in its metadata or search filters.
- 4Editorial teams using AI tools such as Firefly to extend, retouch, or replace elements of a real photograph without disclosing the alteration to readers.
- 5Social media teams republishing AI-generated images that circulate as if they were documentary photographs, without verifying provenance.
- 6Commissioning AI-generated portraits or scene reconstructions for features and failing to label them clearly as illustrations rather than photographs.
- 7Syndicating content between publications where one outlet has stripped C2PA Content Credentials metadata during file conversion.
- 8Using AI upscaling or enhancement tools on archival photographs in ways that materially alter the image beyond the original capture.
Red flags
- An image supplied to the newsroom has no verifiable provenance — no photographer credit, no agency watermark, no C2PA metadata — and depicts a newsworthy event.
- Characteristic AI artefacts are visible: impossible finger counts, inconsistent text within the image, unnaturally smooth skin textures, or implausible background geometry.
- A source sends a photograph that contradicts all other visual evidence from the same scene — a strong indicator of deepfake manipulation.
- The image has been sent as a flattened JPEG with EXIF data stripped, making it impossible to verify camera, location, or timestamp without further investigation.
- An AI-generated image is described in the caption as a photograph, or the AI origin is disclosed only in the article body rather than adjacent to the image itself.
- A freelancer or contributor submits imagery generated by an AI tool without declaring that fact — in breach of NUJ AI guidance and the publication's contributor terms.
AI imagery disclosure checklist
- I have verified the provenance of this image: photographer credit, agency, or C2PA Content Credentials metadata.
- If the image is AI-generated or AI-altered, I have added a visible label in or adjacent to the caption before publication.
- I have checked the image using the Content Credentials Verify tool at contentcredentials.org where credentials are available.
- I have run a reverse image search (Google Images or TinEye) for images from unfamiliar sources claiming to depict real events.
- The caption and alt text both reflect the AI-generated or AI-illustrated nature of the image where applicable.
- I have confirmed the image does not depict a real identifiable person in a fabricated context that could be defamatory or misleading.
- The publication's AI imagery policy has been followed, and the picture editor or relevant senior editor has approved use of AI-generated visuals.
- Where the image is illustrative (not documentary), it is clearly labelled as such and is not positioned to suggest it depicts a real event.
AI content ethics guidance
Our AI-Generated Content ethics hub covers text as well as images, including disclosure obligations, editorial review workflows, and the NUJ's position on AI use across all content types.
Common mistakes
- Labelling AI-generated imagery only in the article body and not adjacent to the image — readers who scan visuals without reading the text will miss the disclosure entirely.
- Treating an AI “photorealistic illustration” as equivalent to a commissioned artist's graphic — without clear labelling, readers cannot distinguish it from a genuine photograph.
- Assuming that because an AI tool (such as Adobe Firefly) is “trained on licensed content” the output is automatically safe to publish without further editorial consideration.
- Stripping image metadata during export or CMS upload, inadvertently removing C2PA Content Credentials that would have supported provenance verification.
- Applying AI-based background replacement or object removal to a documentary photograph without disclosing the alteration — this materially changes the factual record of the image.
- Failing to update contributor agreements to require disclosure when freelancers use AI tools to create or alter submitted imagery.
Related guides
Primary sources
- IPSO Editors' Code of Practice— IPSO
- NUJ AI Guidance— NUJ
- Content Credentials — C2PA— C2PA
- Reuters Trust Principles— Reuters
- Adobe Firefly— Adobe
- Journalism.co.uk— Journalism.co.uk
Frequently asked questions
What does IPSO's Editors' Code say about AI-generated images?
IPSO's Editors' Code of Practice Clause 1 (Accuracy) requires that publications not publish inaccurate, misleading, or distorted information. The Code does not yet contain an explicit AI-generated imagery clause, but IPSO's published guidance makes clear that using AI-generated images in a way that misleads readers about events or people — particularly without disclosure — risks a Clause 1 breach. Editors are expected to take care over any content, including visuals, that could create a false impression of fact.
What is 'significant AI use' and when must it be disclosed?
The NUJ's AI guidance describes 'significant AI use' as situations where AI has materially shaped the content a reader sees — for example, an image generated entirely by an AI tool rather than captured by a photographer. The NUJ recommends that members disclose when AI tools have been used to create or substantially alter an image, and that the disclosure be prominent enough that readers can make an informed judgement about the content. The Reuters Trust Principles similarly require transparency about how content is produced.
What are C2PA Content Credentials and how do they help?
The Coalition for Content Provenance and Authenticity (C2PA) has developed a technical standard called Content Credentials that embeds tamper-evident metadata into image files, recording how and where an image was created or modified. Tools such as Adobe Firefly and certain camera manufacturers now attach C2PA credentials by default. Journalists and editors can verify an image's provenance using the Content Credentials Verify tool at contentcredentials.org. This does not replace editorial disclosure to readers, but it supports internal verification workflows.
Is there a difference between editorial and illustrative use of AI imagery?
Yes. Editorial use refers to images used to represent or report on real events or real people. Using an AI-generated image in an editorial context to illustrate a story about a real event or real person is particularly risky: it can create a false impression of fact, which is a Clause 1 concern under the IPSO Editors' Code and may also engage defamation risks. Illustrative use — clearly labelled conceptual or decorative imagery that does not purport to depict reality — is lower risk, provided the labelling is unambiguous and prominent.
How can journalists detect AI-generated or manipulated images?
Detection is imperfect and evolving. Practical approaches include: checking C2PA Content Credentials metadata using contentcredentials.org; running reverse image searches via Google Images or TinEye to check whether a claimed documentary photograph has a verifiable origin; examining the image for characteristic AI artefacts (distorted hands, inconsistent backgrounds, implausible text within the image); and using detection tools such as those offered by Hive Moderation or Illuminarty, while noting that no detector is conclusive. Internal verification workflows and editorial judgement remain essential.
What labelling conventions should UK newsrooms adopt for AI imagery?
There is no single mandated UK standard, but the NUJ and leading editorial organisations recommend that AI-generated or AI-altered images carry a visible label in or adjacent to the caption — for example, 'Image: AI-generated illustration' or 'Photograph digitally altered using AI.' The label should be positioned so it is seen before the reader engages with the image, not buried in small print at the foot of the page. Alt text for accessibility should also reflect the AI-generated nature of the image.