Artificial Intelligence Policy

Talon’s AI Policy

Why are these policies being set?

AI has the tendency to hallucinate in ways that can harm the journalistic integrity and accuracy of The Talon. Generative AI can take from copyrighted material in ways that are potentially infringement. The substantial use of AI by a newspaper can erode trust in the paper; simultaneously, AI has the power to speed up our work by automating tedious activities. This policy is designed to encourage positive use of AI while discouraging negative use. 

What is the general policy?

The Talon will not use AI to generate writing. All AI content must be overseen by humans and all use of AI must be approved before use. AI is never the final editor of anything that we publish. AI should never be used to generate facts or fact-check information. All use of generative AI should be disclosed. If a piece of information included in an article due to AI generation is found to be false, misattributed, etc., the author(s) of the article will be held responsible as if they wrote the material. 

What are the allowed uses of AI?

Transcribing audio: AI can be used to summarize or transcribe recorded interviews, which does not have to be disclosed in our reporting. However, humans must review the audio to ensure accurate transcription. If an unfaithful transcription is used, the author(s) of the article will be held responsible as if they wrote the material.

Translation: Final translated text must be edited by a human who is fluent in all involved languages. AI translation is allowed for initial understanding, but we’re encouraged to have humans fully translate text. Use of AI translation must be disclosed in the article (e.g. “Initial translation was done by [AI platform: Google Translate]. The translation was edited/reviewed/verified by [human translator]”).

Editing: AI [e.g. Grammarly (not Grammarly AI), Murrow] may be used for light copy-editing such as spelling and grammar-checking, given that a human is still the final editor of the piece. This does not need to be disclosed. Substantial edits (e.g. shortening/lengthening an article) must be made by humans. 

Audio: Generic text-to-speech AI voices can be used when human narration is impossible (i.e. an anonymous source doesn’t want their voice to be published, and voice modulation/other tactics aren’t possible). Disclosure rules apply. AI will not be used to replicate the voice of any person. 

Media: We will not use AI-generated media (images, videos, etc.), and even in exceptional circumstances, we will never publish any AI images that could easily pass for photographs of real-world events. Photo editing tools that utilize AI (e.g. AI Denoise in Lightroom) are allowed as long as they don’t change any substantial facts or implications communicated by the photo (e.g. changing the lighting to make it look like a different time of day, garbling text in a corner of the photo). 

Are there exceptions?

Exceptions to these rules can be made in rare circumstances, such as using GenAI content when writing an article about generative AI. Even if exceptions are made, all use must be clearly disclosed. 

Future-proofing: This is a living document, and it will change as the AI landscape changes. All feedback about The Talon is welcome at [email protected]. 

If you have any questions about anything in this document, please talk to your editor and/or Oscar Johnson (WME) Dorie Xie (PME) and/or Milan Grbovic (EIC).