Article translated by Owen Huchon — Original version published on the Nieman Lab website on July 11, 2023.
The emergence of generative AI has highlighted the need to establish guidelines for the use of these technologies in newsrooms. In this post, we'll examine a sample of guidelines implemented by media organizations.
In the first section, we will outline certain themes and patterns of varying importance that we observed. In the second section, and based on our analysis, we suggest some approaches that may prove useful in developing codes of conduct. Whether you are a curious journalist or someone responsible in a newsroom, we hope this document will serve as an overview of potential safeguards for generative AI.
The selection of rules we analyzed covers a range of large and smaller organizations, primarily in Europe and the United States, but also in other parts of the world. The current sample of 21 guidelines can be found here.
The elements we analyzed vary in specificity and are sometimes named differently as "editor's note," "protocol," "principles," or even "code of ethics." The tone of certain guidelines is restrictive and prohibits specific uses. Other documents are more management examples, in which news organizations commit to assuming specific responsibilities to make AI less risky. We will examine below some of these general patterns using examples drawn from specific guidelines.
Observations from published documents:
SUPERVISION
Some rules mention the concept of oversight and link it to the importance of meaningful human participation and monitoring in the use of AI, notably through additional review and fact-checking before publication. News organizations also reject the idea of replacing journalists with machines and emphasize the importance of humans' decision-making role when using generative AI tools.
Aftonbladet and VG are Swedish and Norwegian news outlets. Both owned by the press company Schibsted, they adopt similar positions. Aftonbladet's guidelines state that "all published material has been reviewed by a human and is under our authority," while those of VG specify that "any use of generative AI must respect privacy."
Reuters applies a supervision system based on "ensuring meaningful human involvement and human-centered design, development, deployment of AI products and data use in a way that treats people fairly". The Guardian for its part states that the use of generative AI requires human involvement, specifying in its guidelines that its use must be conditional on "a specific benefit and explicit authorization from an editor-in-chief". The ANP, the Dutch news agency, takes a similar position on this issue, indicating that it can use AI or similar systems to "assist with final editing, provided that a human performs a final check afterward". It describes this method as a human → machine → human process, in which decision-making is supervised by a human.
CBC, Canada's public broadcasting service, ensures that "no CBC report will be published or broadcast without direct human intervention and oversight". De Volkskrant, a leading newspaper in the Netherlands, states that its content is created by human editors, reporters, photographers and illustrators, and that AI-generated content cannot be used under the Volkskrant name without approval from an individual. The aforementioned policies of Volkskrant are very similar to those of Heidi.News in Switzerland and Le Parisien in France. Generally speaking, they all assert that no content will be published without prior human oversight.
Nucleo, a Brazilian digital media outlet, stated that it will never publish AI-generated content without human oversight in articles and website posts and that it will not use AI as the final editor or producer of a publication. Ringier, a Swiss media group that owns brands in 19 countries, states that results generated by AI tools must always be subject to critical review and information must be verified, checked and completed using the company's judgment and expertise. Similarly, German news agency DPA clarifies that the final decision regarding AI use is made by a human. It adds: We respect human autonomy and the primacy of human decision-making.
The German Association of Journalists (DJV) refers to AI as "colleagues": "Under no circumstances should AI colleagues be allowed to replace editors-in-chief. Even if the use of artificial intelligence changes or completely eliminates tasks, it does not make the people who make up the editorial team superfluous". The Financial Times affirms that its journalism will continue to be reported, written, and edited by humans who are the best in their field. STT, a Finnish news agency, links monitoring to decision-making: "Don't let AI decide for you, but always evaluate the usefulness of the answers yourself".
Insider, an American online media company, states that its journalists should always verify their facts. They specifically add: "Don't plagiarize! Always check for originality. Best practices in this area are likely to evolve, but for now, make sure at minimum that you submit all passages received from ChatGPT to Google Search and Grammarly's plagiarism check". They also ask editors-in-chief to encourage journalists to fact-check. "It's necessary to increase your vigilance and take time to ask yourself how each fact in every article is known to your fellow journalist".
TRANSPARENCY
Transparency disclosures are manifested through clear labeling of content in a way that is understandable to the public. However, guidelines often fall short of clearly indicating how these transparency disclosures will translate into practice.
Aftonbladet and VG state that in the rare cases where they publish AI-generated material, whether text or images, it will be clearly indicated. VG specifically adds: "AI-generated content must be clearly and comprehensibly labeled". Along the same lines, Reuters states that it will implement practices "to make the use of data and AI in our products and services understandable".
The Guardian states that when it uses generative AI, it will do so transparently. CBC promises "no surprises", committing to label AI-generated content: "We will not use or present AI-generated content to viewers without fully informing them". Similar statements appear in the guidelines of the Dutch news agency (ANP), the news group Mediahuis, the Council for Journalism of the Flemish press (RVDJ), the German news agency (DPA) and the Association of German Journalists (DJV), all stating that AI-generated content must be clearly labeled.
Ringier, a Swiss media group, states that AI-generated content must be labeled. However, this requirement is not necessary « in cases where AI is used only as an aid », which suggests a different approach to transparency when AI is used by people as an integral part of the workflow. The STT indicates that AI use in reporting must always be communicated to the public: « This applies both to situations where the technology was used to help produce stories and to content whose source material was created by machine intelligence ».
PROHIBITED USES VS. PERMITTED USES
While these guides generally list the prohibited and permitted uses of generative AI, they are sometimes also accompanied by exceptions. Conditions for transparency, accountability, and reporting requirements are then detailed in the guidelines. Image generation use cases receive particular attention.
Wired states it does not publish articles containing AI-generated text, except when the AI-generated character is the article's central element. It also specifies that it "will not publish AI-edited text" and that it "will not use AI-generated images in place of stock photos". The underlying reasoning suggests that editing is inherently tied to determining what is relevant, original, and engaging. Despite the restrictive tone of Wired's guidelines, the company states it may experiment with AI for tasks related to the reporting process, such as suggesting headlines or generating ideas or copy for short social media posts.
Nucleo lists certain permitted uses of generative AI, such as summarizing texts to suggest alternative social media posts and as a research tool on topics and themes. Journalists at Aftonbladet and VG, for example, can use AI technology in their work to produce illustrations, graphics and templates. VG's guidelines, however, add a prohibition on generating photorealistic images, because AI-generated content must never undermine the credibility of journalistic photography.
Insider states that its journalists can use generative AI for "article drafts, SEO headlines, editing, interview questions, concept explanations and summaries of past reporting". However, they clarify that their journalists must not use it to write on their behalf. Regarding the use of AI-generated images, they emphasize that they must "have more discussions before deciding whether and how to use them".
Hong Kong Free Press (HKFP), an English-language news outlet in Hong Kong, restricts all use of generative AI for article writing, image generation, or fact-checking, because "few AI tools include proper source information or attribution". Notably, the HKFP states it may use internally approved AI "for grammar and spelling checks, for rewriting and summarizing existing texts written by our team, and for help with translation, transcription, or research". Similarly, De Volkskrant, a quality newspaper in the Netherlands, states that its journalists do not publish journalistic work generated by artificial intelligence.
CBC states it will not use AI to recreate the voice or image of a CBC journalist or personality except to illustrate how the technology works. It ties this exception to two conditions: (1) prior approval from our standards office and (2) approval from the person being "recreated". They will not use the technology for investigative journalism in the form of facial recognition or voice matching, nor will they use it to generate voices for confidential sources whose identity they are trying to protect. They will continue to use practices their audience understands, such as voice modulation, image blur, and silhouette.
Heidi.news assures that it will only use synthetic images for illustrative purposes and not for informational purposes, so as not to create confusion with real events. It also adds that it will not publish any synthetic image that could pass for a photograph, "except for educational purposes when the image in question is already public".
Le Parisien states that it reserves the right to use AI for generating text and images for illustration purposes. It conditions the exceptional use of generative AI on permanent transparency: "we will ensure that we explicitly specify its origin for the reader". They describe the use of AI as an enrichment. "A journalist may use these tools as they would a search engine, but they must always return to their own sources to guarantee the origin of their information". The Financial Times emphasizes that it will not publish photorealistic images generated by AI, but that it will explore the use of AI-enhanced visuals (infographics, diagrams, photos) and will, where applicable, specify this to readers.
STT draws a direct link between prohibited uses of generative AI and its limitations. « STT does not use AI for data mining. The sources used by AI are often unclear, which makes its use in editorial work problematic ». The agency also notes that source reliability and information verification remain essential.
ACCOUNTABILITY AND RESPONSIBILITY
Accountability and responsibility are often mentioned in guidelines in relation to published content as well as values such as accuracy, fairness, originality and transparency. The implementation of accountability measures for the use of data and AI products is also emphasized, as is the use of technically robust and secure AI systems to minimize risks.
Aftonbladet states it is responsible for "everything it publishes on the site, including material produced using or based on AI or other technologies, and which falls within our publishing authority". Reuters states it will "implement and maintain appropriate accountability measures for our use of data and our AI products and services". DPA, the German news agency, emphasizes it will only use AI that is "technically robust and secure to minimize the risk of error and misuse". Le Parisien states it wants to protect itself against "any risk of error, copyright infringement, and also to support the work of artists, photographers and illustrators".
The Association of German Journalists (DJV) stresses that press organizations are responsible for their content and that editorial teams must establish regulated processes for accepting and approving journalistic content when AI is involved. The Financial Times says it is convinced that its "mission to produce journalism of the highest quality is all the more important in this era of technological innovation". It adds: "The FT has a greater responsibility in terms of transparency, presenting facts and seeking the truth".
Insider states that its audience must trust it and hold it accountable for the accuracy, fairness, originality, and quality of every article. They assert that their journalists are responsible for meeting these standards and for the quality of every word in their articles.
PRIVACY AND CONFIDENTIALITY
Rules designed to uphold confidentiality principles often pertain to the protection of sources and the handling of sensitive information on external platforms. In addition, the policies of some media outlets emphasize the need for caution when using confidential or unpublished content as input for generative AI tools.
Aftonbladet states that they “uphold source protection and do not feed sensitive or exclusive information to external platforms such as ChatGPT,” a sentiment also echoed, in slightly different terms, by VG: “Journalists should initially share with AI services only material that has been approved for immediate publication on VG’s platforms.” Reuters emphasizes that it “will prioritize security and privacy in its use of data and throughout the design, development, and deployment of its data and AI products and services.” CBC states that it will not feed confidential or unpublished content into generative AI tools for any reason. Mediahuis states that it must comply with privacy laws and, where applicable, obtain user consent before using personal information.
Ringier states that its employees “are not permitted to enter confidential information, trade secrets, or personal data from journalistic sources, employees, customers, business partners, or other individuals into an AI tool.” With regard to the development of code, Ringier’s guidelines specify that code may only be entered into generative AI systems if it “does not constitute a trade secret and does not belong to third parties.”
It is interesting to note that most of the guidelines we analyzed do not distinguish between generative AI services (e.g., OpenAI) and the development and use of a generative AI system hosted on computers operated by the organization itself, as might be the case with open-source models. It is important to note, however, that privacy and confidentiality risks are more closely tied to the use of generative AI services hosted by other organizations than to the use of generative AI itself.
CAUTIOUS EXPERIMENTATION
References to “cautious experimentation” in the guidelines are often linked to curiosity and critical thinking, while highlighting the potential for innovation. Emphasis is also placed on verifying the accuracy of results generated by AI, as the focus is on recognizing the risks of misinformation and “corruption of the truth.”
VG’s main rule: its journalists must treat “the use of text, video, images, audio, and other content created with the help of AI (generative AI) in the same way as other sources of information—with open curiosity and caution.” The CBC states that it will never rely solely on AI-generated research in its journalism: “We always use multiple sources to confirm facts.” ANP states that every member of the ANP newsroom should approach AI and related systems “with wonder, curiosity, a critical mindset, and openness to new developments.” The agency adds that it continues to actively seek out innovations with an open mind. STT states that journalists should explore AI whenever they have the opportunity. It was one of the first news publishers to begin experimenting with generative AI in the form of language models, and asserts that experimentation remains essential to discovering the possibilities and limitations of these models.
The Financial Times states in its letter of intent that it will adopt AI to provide services to its readers and customers. It is interesting to note that it also emphasizes the need for the FT to establish a team in the newsroom that can experiment with AI responsibly. They add that “every technology opens up exciting new frontiers that must be explored responsibly. But as recent history has shown, enthusiasm must be tempered by a degree of caution in the face of the risk of misinformation and the distortion of the truth.” Finally, the FT states that all AI experiments in newsrooms will be recorded in an internal log, including, to the extent possible, the use of third-party providers who may utilize the tool.
STRATEGIC INTENT FOR USE
In many cases, policy documents also outline the organization’s strategic objectives regarding the deployment of generative AI. The motivations cited include a desire to improve originality, quality, and even speed, while at times reflecting a commitment to not replace journalists and to uphold core values such as independence and impartiality.
The Financial Times states that AI has the potential “to increase their productivity and free up time for journalists and editors to focus on producing and presenting original content.” The Guardian states that it will also strive to use generative AI tools in editorial work only when they contribute to the creation and dissemination of original journalism. Furthermore, it specifies that when using generative AI, it will focus on “situations where it can improve the quality of our work.” Mediahuis states that AI should improve its journalism. Its goal is to “improve the quality of our journalism for our audience.” Insider emphasizes that its journalists can—and indeed should—use AI to improve their work: “But this remains your work,” it states, “and you are responsible for it. You are accountable to our readers and viewers.” The ANP, the Dutch news agency, states that it wants to remain impartial and independent and that it should exercise caution in the use of generative AI.
The DPA, the German news agency, states that it uses AI for various purposes and is open to increased use of this tool. It adds that AI will help it improve and speed up its work. De Volkskrant, a quality newspaper in the Netherlands, views artificial intelligence (AI) as a tool, “never as a system capable of replacing the work of a journalist,” and this sentiment—that journalists should not be replaced by machines—is shared by Heidi.News and Le Parisien. Heidi.News believes that AI systems should be used as tools and that, in the interest of objectivity, its journalists must under no circumstances consider them “sources of information.”
TRAINING
Training is rarely mentioned in the guidelines, but when it is, the training programs and courses primarily focus on mitigating the risks of generative AI and on accountability and transparency toward the public.
Mediahuis asserts that training and certification should be established for those responsible for AI-related decisions. They link this to the development of clear lines of accountability for the development and use of AI. The German Journalists’ Association (DJV) emphasizes that the use of artificial intelligence must be an integral part of journalists’ training and professional development. It calls on media companies to implement appropriate training programs that address the misuse of AI. The British newspaper The Financial Times states that it will offer its journalists training on the use of generative AI for story discovery, which will be provided in the form of masterclasses.
BIAS
The Guardian explicitly addresses bias in generative AI and states that it “will guard against the dangers of bias embedded in generative tools and their underlying training datasets.” Mediahuis states that its journalists must be mindful of biases in AI systems and strive to address them. Ringier states that its tools must always be “fair, impartial, and non-discriminatory.”
FLEXIBILITY OF THE GUIDELINES
Several documents have demonstrated humility in the face of rapid change and have emphasized the importance of adapting these policies over time, as our understanding of the risks evolves.
Nucleo states that its policy on generative AI is “constantly evolving and may be updated from time to time.” A similar statement can be found in the guidelines of Aftonbladet and VG. Wired writes that AI will continue to develop, which “may change our perspective over time, and we will note any changes in this post.” According to an editor’s note at the bottom of the document, these guidelines have already been updated once. The CBC states that it has informed its journalists that these rules are “preliminary and subject to change based on technological developments and industry best practices.” De Volkskrant states that it is following developments with a critical eye and will adapt its protocol if it deems it necessary. Similarly, Ringier states that its policies will be continuously reviewed in the coming months and adjusted as necessary.
Topics that are less frequently mentioned in the guidelines:
Legal compliance, as well as customization, data quality, user feedback, and the integration of generative AI into the supply chain, are rarely mentioned.
SUPPLY CHAIN
The AI supply chain consists of a network of suppliers for AI systems, such as third-party models, collaborators, data providers, annotation providers, etc. The German Journalists’ Association (DJV) emphasizes the need for precision in data processing. The association stresses the importance of accuracy in data processing: “The collection, preparation, and processing of data must meet high quality standards. Incomplete data, distortions, and other errors must be corrected without delay.” It is interesting to note that the DJV encourages media organizations to create their own value-based databases and supports open data projects led by public authorities and government institutions. They assert that greater independence from commercial big-tech providers is desirable. Reuters is one of the few news organizations to mention the governance of collaborations, stating that “they will strive to partner with individuals and organizations that share ethical approaches similar to ours regarding the use of data, content, and AI.”
LEGAL COMPLIANCE
The DPA states that it uses AI only in compliance with applicable laws and legal requirements and that it adheres to its ethical principles, such as human autonomy, fairness, and democratic values. The Association of German Journalists is calling on lawmakers to ensure that the labeling of AI-generated content becomes mandatory. It is urging them to enshrine these labeling requirements in law.
CUSTOMIZATION
The same organization states that the use of AI to personalize services must always be carried out in a responsible and balanced manner. It adds that users should have “the option to modify the selection criteria and/or completely disable personalized recommendations.”
USER FEEDBACK
Mediahuis encourages its readers to share their feedback and allow the company to review their data. This reflects a commitment to transparency, accountability, and a customer-centric mindset in the evolving landscape of media consumption.
A Few Guidelines for the Guidelines
Based on the observations above, we offer here a few suggestions for journalists and news organizations considering the development of their own codes of ethics. By reviewing existing guidelines and codes of ethics, adopting a risk-based approach, and drawing on diverse perspectives within the organization, we believe that news organizations can improve the quality and effectiveness of their rules regarding generative AI.
REVIEW EXISTING GUIDELINES AND CODES OF ETHICS
Several themes addressed in our analysis—including accountability, transparency, privacy protection, and so on—reflect well-established values and ethical principles in journalistic practice. Therefore, when developing or updating guidelines in light of generative AI, we suggest that it would be useful to examine existing codes of conduct and journalistic principles to consider how these principles can be upheld in the face of the changes brought about by generative AI. For example, although the idea of independence appears several times in the documents we analyzed, it is not as prominent as one might expect, given that it is central, for instance, in codes of ethics such as that of the SPJ. How might the use of generative AI fully embody and reflect a normative commitment to journalistic independence?
We emphasize that the emergence of generative AI may present new challenges for news organizations, but that does not mean that the development of guidelines must start from scratch. We suggest that news organizations review these codes of conduct and evaluate them—one by one—in light of the potential use of generative AI and the risks it may pose. By systematically examining the core values and principles of journalism, they should propose strategies and tactics for using generative AI in a manner consistent with established standards.
ADOPT A RISK ASSESSMENT APPROACH.
Media organizations can benefit from a systematic approach to risk assessment when developing guidelines. Such a method, as developed by NIST, can, for example, help map, measure, and manage risks by providing tools to effectively develop policies to mitigate identified risks. What are the various risks associated with the use of generative AI, how can these risks be measured and monitored, and how can they be managed in a way that aligns with a news organization’s existing values and objectives?
For example, traditional journalistic practices emphasize the importance of relying on credible sources and ensuring the accuracy of information. However, generative AI introduces a degree of uncertainty regarding the origin and reliability of content that may be obtained during the reporting process. One way to mitigate these risks could be to establish procedures for verifying the authenticity of documents shared as sources. Journalists could address these uncertainties by implementing rigorous verification processes and using corroboration and triangulation techniques to validate AI-generated content. By cross-checking information from multiple sources, newsrooms could reduce the risk of publishing unverified or false information.
CREATE A DIVERSE GROUP TO DRAFT GUIDELINES.
Although generative AI can also be used to generate computer code—and can therefore have an impact on all sorts of different information products—we have found few guidelines that distinguish between the various possible approaches for editorial use and the use of products (for example, in software development). Furthermore, other areas of news organizations, such as sales and marketing, could also benefit from the use of generative AI. Should they have separate guidelines? To cover the entire scope, we suggest that it is important to bring together a diverse group of stakeholders within your media organization to discuss the guidelines and define their scope appropriately. This will allow organizations to consider the risks that could arise in the newsroom, and it could help identify broader, company-wide risks that go beyond journalists’ day-to-day workflows.
Editor's note
This text is not a word-for-word translation of the original article published on the Nieman Lab website. We have added contextual information, adapted certain expressions, and omitted certain passages containing specialized references intended for English-speaking or American readers. We have also chosen to apply the rules of inclusive writing to this translation, in accordance with the editorial guidelines currently in effect for all Médianes content.
