Our advice about AI transparency is evolving
You can see when this advice was last updated above – please check back regularly.
The BBC is committed to being open and transparent about our use of AI. This is underpinned by our BBC AI Principles, including Transparency and Clear Explanations:
We will be clear with audiences and staff about where we use AI and what data we collect. We will make sure they can understand why we use it, how it works and how it affects them.
This advice outlines key principles to follow to achieve this. Specific use cases and examples included reflect a broad range of wider industry use cases. They do not necessarily indicate where the BBC is using or planning to use AI.
Before considering transparency, you should confirm with your commissioner or key BBC contact that any proposed use of AI is aligned with the BBC’s policies and guidance.
Why transparency matters
AI-generated content can blur the lines between what is real or artificial. It’s important for us to help audiences and others understand the role AI plays in our work, and how it impacts the content they consume. Being transparent about where and when we use AI helps to maintain trust in our content and services.
Audience demand for AI transparency is driven by 4 key needs. Our audiences have told us they want to be:
- reassured that AI is used as an assistive tool, and that creativity and journalistic instinct are preserved
- empowered to evaluate content effectively
- informed more broadly about AI and what it means for media, and
- respected by being shown how the BBC is using AI
These needs vary across types of content and will evolve over time. They are heightened when it comes to news, current affairs, or factual content where audiences fear they could be misled.
Transparency is also important to create a culture of openness – with staff, suppliers, contributors, and partners – to ensure accountability around the use of AI. Being transparent helps everyone make informed decisions.
How to be transparent
Disclosure of AI use is the main way to achieve transparency. There are 2 main types of disclosure – direct and indirect disclosure:
|
|
What it involves |
Example |
|---|---|---|
|
Direct disclosure |
Labelling or otherwise revealing the use of AI at the point of consumption |
Annotating an image as AI generated |
|
Indirect disclosure |
Labelling or other interventions not present at the point of consumption |
Disclosing in content credits, or website terms and conditions |
When to disclose the use of AI in content for audiences
What the Editorial Guidance on AI says:
Any use of AI by the BBC in the creation, presentation or distribution of content must be transparent and clear to the audience. The audience should be informed in a manner appropriate to the context, and it may be helpful to explain not just that AI has been used but how and why it has been used.
Direct disclosure is required if either of the following apply:
- the use of AI risks materially misleading audiences. This is to ensure that artificial output that could be mistaken for real output is clearly labelled to disclose the role of AI. For example, the use of AI “face-swapping” techniques to anonymize contributors.
- AI is used to automate output without the direct oversight or review of a human. For example, using AI to automatically generate subtitles for podcasts.
News, current affairs, and factual content
Generative AI should not be used directly for creating content for News, Current Affairs or Factual content. Exemptions may be considered, subject to a piloting process. In those cases, you should provide direct disclosure as required above, and also if:
- AI has a material or significant impact on the content or in its production – think about how much was this created by a journalist, and how much was created by AI. For example, if an AI transcript from a press conference is used for research, you do not need to disclose. But if a whole article has been translated with the assistance of AI, you should.
If you’re an external producer…
External producers should discuss any use of AI which might require disclosure with your commissioner, who will consult Editorial Policy as appropriate.
See below for guidance on how to disclose.
When to disclose the use of AI for anything else (not content for audiences)
Where AI is used for anything other than audience facing content, the broad principles above should guide the need for disclosure: risk of materially misleading, use of automation, and significance of impact.
Producers and suppliers should discuss proposed uses of AI with your commissioner or key BBC contact, who will consult relevant internal teams to determine disclosure requirements.
How to disclose use of AI
Where direct or indirect disclosure of the use of AI is required, our audiences expect clarity about the nature of the roles played by humans. This includes the level of oversight of AI in the process.
For clarity and consistency, you should use the following formula as a broad guide to key decisions when drafting disclosure. You can adapt it to suit the nature and tone of your content. Please discuss your draft approach with your commissioner or key BBC contact, who will consult as appropriate to agree final wording.
Disclosure Formula
Content format + Role of human (including oversight) + Role of AI + Link for more information
- Content format is a description of the content.
- Role of human (including oversight) makes human involvement and control clear, particularly human oversight and review, or being clear where AI is operating automatically.
- Role of AI clarifies what AI is doing. It’s important to stress its assistive nature where relevant.
- Link to more information – if technically possible you can link to more about your specific use of AI (e.g. a blog post or external communications)
Disclosing AI alongside other types of disclosure
Sometimes AI might be used in content that already requires disclosure. For example, using AI to help generate a reconstruction of a factual event. In these cases, you don’t need to separately disclose AI – you only need one single disclosure. However, if your use of AI is something that might materially contribute to any risk of misleading audiences, it should be included in the disclosure.
Examples
Here are some examples of how to apply the Disclosure Formula. As we build a library of consistent ways to achieve disclosure we will add more examples. As above, if your use case isn’t covered, please get in touch for advice.
- This article was written and reviewed by a journalist, using AI to assist with translation. More on how the BBC uses AI [LINK]
- Today’s match is being delivered to you by (name of journalist) with the help of Al tools which transcribe and summarise BBC radio coverage. Our ambition is to use these tools to allow us to produce live pages for games that we otherwise wouldn’t be able to cover, and this trial will help us develop that technology. All posts will be checked by (name of editor) before publishing.
