At first glance, many media companies have tackled the subject of AI in an almost exemplary manner. They got off to a flying start, experimented, formed interdisciplinary teams, appointed AI directors and – after all, they do have a responsibility – developed ethical guidelines. Of course, the industry was also able to express outrage at a few moral outliers – “Sports Illustrated!”, “the Michael Schumacher interview!”, “Burda’s cookery magazine!” – but in many places, lengthy lists of “dos and don’ts” were intended to prevent the worst. It’s just a pity that most of these sets of rules are likely to prove, before long, to be a form of AI-washing. For they give the illusion of control that has long since slipped from the media companies’ grasp. To paraphrase a line from Rainer Maria Rilke‘s famous poem „Autumn Day“: Whoever has no moral compass now will never build one.
The long version is this: rapid technological progress and the power of tech conglomerates, coupled with economic and, in some places, political pressure, have created realities that even exemplary management finds difficult to cope with. This is suggested by research for the EBU News Report “Leading Newsrooms in the Age of Generative AI“, published in 2025.
First and foremost, there is the issue of ‘shadow AI’. One media manager observes that the biggest changes are not currently being driven by media organisations, but are arising simply because journalists are using AI tools. Unlike 25 years ago, when editors and reporters still had to be painstakingly convinced of the value of digital tools and platforms, AI tools are so intuitive to use that people employ them even more frequently in their private lives than at work, as a recently published study found – at least with regard to the US. A report by the Thomson Reuters Foundation supports this view for journalists in the Global South: 80 per cent of those surveyed used AI at work, but not even 20 per cent of their newsrooms had a corresponding strategy or policy in place. Yet within organisations, both groups can become a problem: the tech-savvy staff who overstep boundaries whilst experimenting, and the less tech-savvy staff who make mistakes out of ignorance, such as disclosing sensitive data.
Added to this is the fact that many ethical guidelines are not practical for day-to-day use. The BBC is no doubt proud that it has condensed its current guidance on AI into just nine points – plus sub-points. But one cannot expect editors, who are under such time pressure during their shifts that some do not even dare to go to the loo, to have internalised all the regulations. The workload is likely to push journalists even more towards using AI. Just as text from dpa reports has occasionally had to be used in the past, an LLM will be consulted in future if it saves time.
The rule that is currently most widely followed is particularly difficult to adhere to: ‘Human in the loop’ – a human should have the final say on AI-generated content before it is published . Even in day-to-day operations, editors overlook errors. When AI tools multiply the speed of output, humans reach their cognitive limits. And if they were to work meticulously nonetheless, they would inevitably prevent the efficiency gains hoped for by management. The ‘human in the loop’ principle undermines the scalability that is expected from AI, writes Felix Simon in a commentary for the Reuters Institute.
When in doubt, editorial teams wriggle out of the requirement with disclaimers. They state that content translated or produced using AI is labelled as such – in other words: they accept no responsibility for errors. This can work well and is accepted by the audience in some cases, for example with subtitles for TV programmes. Here, the desire for comprehensibility takes precedence, for instance for the hearing-impaired. However, it can also produce rubbish, as with articles from the Washington Post that are translated by AI and published in the Ippen Group’s publications. Furthermore, even in high-quality journalism, there are scenarios where rigid rules are of no help. If, for example, cloned voices were banned across the board, it would limit narrative possibilities. Even public service broadcasters have used voice clones of historical figures to bring contemporary history to life.
In all these cases, guidelines tucked away on the intranet are of no help. A more effective approach is a mix of technical solutions, training and debate: desired applications must be automated within the CMS. Through experiments and training sessions, users can acquire knowledge of AI and learn how to work with it. And those who regularly discuss values – even in highly contentious cases – are more likely to reflect on them and act accordingly. Monitoring staff at every stage of research and production has never worked in journalism. Each individual must calibrate their own moral compass and follow it.
However, this is of little use if the management doesn’t have one. The owner of the Los Angeles Times, for example, recently ordered the editorial team to tag comments using a tool called the Bias Meter. The AI automatically alerts readers to opposing viewpoints. Clearly, the output isn’t being proofread by humans. Otherwise, someone might well have noticed that the machine had cast the Ku Klux Klan in a somewhat too favourable light. But those are details. No AI tool will ever be able to replace a lack of trust. Yet using AI in this way can destroy hard-won trust – both amongst staff and with the public.
What could, however, render ethical guidelines in media organisations entirely obsolete is the dominance of tech giants. The more AI is embedded in all the tools that everyone uses in their daily lives, the less users will question the values underlying them. The fact that a smartphone camera always makes the sky a little bluer than the one you see in front of you – fair enough. The fact that search engines make answers increasingly easy to digest thanks to AI – that’s fine. The fact that word processing programmes are increasingly acting as editors before an editor has even seen the piece – why not? Only the major organisations will be able to afford to embed their own standards in their systems, and even those may well be infiltrated by AI. The rest will be working with Office and the like.
None of this has to be a bad thing. Autopilots have made air travel many times safer; autonomous driving will achieve the same on the roads. Perhaps AI-assisted journalism will also manage to raise the standard of overall output significantly. After the devastation of the ‘reach’ era, that’s not such a difficult task for some publications. AI tools may even eventually help to implement ethical guidelines. The crucial question is whose rules these will be.
This column was published by the German industry publication Medieninsider in German on 17 March 2025.#