Introduction
I return to a theme that continues to stir debate in my workplace: AI, generative AI (genAI), and parliamentary reporting (see Eras, 2025). In this article, I examine existing and emerging patterns in the realm of AI and deepfakes, reflect on AI adoption as a driver of job displacement—including within parliamentary reporting—and conclude by outlining the steps the Parliamentary Reporting Office (PRO) is taking within the Dutch national context.
Politics and AI
AI is still the talk of the town, at home and at the shop, so what is the current status of AI and, for instance, political deepfakes? A number of patterns occur. First, according to the New York Times, fake photos and videos have begun to have an impact on election results, for example in Romania’s presidential elections, which have been held again as a result (Myers and Thompson, 2025).
Secondly, political leaders—those in charge—use AI and deepfakes as well: for fun, such as President Trump posing as the next Pope in a post on Truth Social; or less innocently, such as Trump having fun with a video on Truth Social of former President Obama getting arrested in the White House. In Dutch politics, the biggest party in parliament, PVV (Party for Freedom), is using AI, probably ChatGPT, to generate pictures of an all-white Netherlands. In one example, in a post on X, a PVV voter, depicted as a cheerful white woman, is contrasted with a voter for the Labour Party (PvdA), depicted as an angry-looking Muslim woman.
Thirdly, more and more politicians say they use AI in their work, such as the Swedish Prime Minister, Ulf Kristersson, who confessed to consulting AI tools for a second opinion (Bryant, 2025), and the UK Technology Secretary, Peter Kyle, who regularly asks ChatGPT for advice (Stacey, 2025).
Fourthly, the Danish Government propose to give people copyright to their facial features. This is meant to be a powerful tool for individuals, including politicians, to remove deepfakes from the internet (Bryant, 2025).
All these patterns indicate that the use of deepfakes in politics, and the use of AI by politicians, is increasingly becoming mainstream.
AI and potential loss of jobs
The potential loss of jobs as a result of the introduction of AI and genAI remains a topic that concerns everyone working in reporting. In July this year, Microsoft came up with another job at risk list, revealing the 40 jobs most exposed to AI (Fortune, 2025). High on the list are the usual suspects, including interpreters and translators (1), writers and authors (5), proofreaders (19) and editors (21). Of course, this is not based on scientific research or data; it is a wish list of a company that has invested $80 billion in AI data centres in 2025 and hopes to generate as much profit as possible (CNN, 2025).
So, what does the data suggest about AI replacing jobs? Exploding Topics (Howarth, 2025) has collected all the stats on AI replacing jobs. Some findings are as expected: AI may replace 300 million jobs worldwide, and companies with annual revenue of at least $500 million are adopting AI more quickly than smaller organisations. More remarkable is that research finds that 81% of office workers are positive about AI and a staggering 80% to 95% of all AI projects fail. The first finding is supported by a British pilot, in which 1,000 civil servants worked with Copilot. This identified high levels of satisfaction from users, particularly from those who are neurodiverse, those who are non-native English speakers, and those with hearing or vision disabilities (Department for Business & Trade, 2025).
The second finding is based on research by both RAND (Ryseff, De Bruhl, Newberry (2024) and MIT (MIT NANDA, 2025). Why do so many AI projects fail? Key causes may be identified as poor data quality, unclear business cases and overestimation of technological capabilities, and may be driven by FOMO (fear of missing out) in companies and organisations. Interestingly, amid all the attention for the benefits of AI and genAI, the costs are often overlooked. In real life, applying AI is a business decision and needs a good use case (Koopman, 2025; DataNorth, 2025). AI costs can be huge, considering the sort of costs that come with implementation. It means buying licences, buying hardware or a platform, cleaning up data, organising maintenance and support, and organising training and compliance (security!).
AI, parliamentary reporting and staff size
To get a clearer picture on this theme, I combined data from three sources: the European Center for Parliamentary Research and Documentation; the IPU Centre for Innovation in Parliament; and the Secretary General de Senado in Spain (ECPRD, 2024; IPU, 2025; Blanco, 2025). The combined data suggests that only a small minority of parliaments and senates actually use AI or genAI for or in parliamentary reporting. The list names legislatures in the following countries: Austria, Belgium, Canada, Estonia, Finland, Germany, Iceland, Latvia, Luxemburg, Portugal, Slovenia and Spain. The effects on staff size—relevant in the light of the likely displacement effect of AI adoption—are inconclusive, with data available only for Luxemburg, Portugal, and Spain. However, the Secretary General de Senado in Spain concludes its report as follows: “The implementation of AI in parliaments is either an ongoing project … or a future project. Although some have advanced in its use, many still rely on traditional methods due to technical problems and the quality of the results. Even some parliaments that state they are very satisfied with the results of using AI for transcriptions do not seem to have reorganised their team or changed their working method”. (Blanco ibid.)
Applying AI at the Dutch Parliamentary Reporting Office
The PRO of the House of Representatives of the Netherlands is also actively exploring the potential adoption of AI in its operations. The Dutch context is severely limited due to a formal prohibition on the use of genAI by the national government and its civil servants, the exception being pilot projects allowed under strict conditions. The PRO started an AI pilot in anticipation of possible personnel and capacity challenges due to an ageing population and labour market developments over the next five to 10 years. The focus of this pilot is therefore on supporting parliamentary reporters, now and in the near future.
Using the ASR tool Whisper and ChatGPT to create our own software, this five-man, three-month pilot aims:
1. to optimise Whisper, using ChatGPT as a Python code developer; and
2. to use ChatGPT as a translator of Whisper text results to produce Hansard-like transcripts.
The premise was to develop a Python interface for text transcription, so that the Whisper text output will be automatically cleaned up or filtered as much as possible in accordance with the conventions of the PRO language guide and editorial standards (for example, standard notation for numbers and dates). The result was called Vluister, and it transcribed a five-minute audio fragment into text in just three minutes. After automatic correction, the output matches the final (official Hansard) product at 80% word level accuracy.
Gaining such experience is crucial for the PRO. Even if the ban on AI is lifted, there will still be no true freedom of choice within the Dutch context, since a European tender procedure remains mandatory. In such cases, subject matter expertise is decisive for the contracting authority—especially considering that big tech shows little interest in parliaments, due to limited budgets and complex demands. This may be another factor in the slow adoption of AI in parliamentary reporting.
Conclusions
As shown above, use of deepfakes in politics and of AI by politicians is increasingly becoming mainstream. The potential for AI adoption as a driver for job displacement looks significant. For parliamentary reporting, however, there is insufficient data to draw conclusions in quantitative terms. For now, there appears to be no major impact in terms of either job losses or changes in working methods. In the Netherlands, the ban on AI for the national government and civil servants makes official experimentation with AI critical for the PRO as the only way to gain relevant experience ahead of a potential procurement process.
Henk-Jan Eras is a Quality Officer with the Parliamentary Reporting Office of the House of Representatives of the Netherlands.
References
Blanco, E. (2025). Note on the Application of Artificial Intelligence in the Transcription and Editing of Parliamentary Debates. Direccion de Asistencia tecnico – Parlementaria Diario de Sesiones, Senado de España.
Bryant, M. (2025). Denmark to tackle deepfakes by giving people copyright to their own features. Guardian. URL: https://www.theguardian.com/technology/2025/jun/27/deepfakes-denmark-copyright-law-artificial-intelligence?CMP=Share_iOSApp_Othe
Bryant, M. (2025). “We didn’t vote for ChatGPT”: Swedish PM under fire for using AI in role. Guardian. URL: ‘We didn’t vote for ChatGPT’: Swedish PM under fire for using AI in role | Artificial intelligence (AI) | The Guardian
CNN, 2025. Microsoft plans to invest $80 billion on AI-enabled data centers in fiscal 2025. CNN. URL: Microsoft plans to invest $80 billion on AI-enabled data centers in fiscal 2025 | CNN Business
DataNorth (2025). AI kosten ontrafeld. URL: Wat zijn de kosten van AI? – DataNorth AI
Department for Business & Trade (2025). The Evaluation of the M365 Copilot Pilot in the Department for Business and Trade. URL: The Evaluation of the M365 Copilot Pilot in the Department for Business and Trade
ECPRD (2024). European Center for Parliamentary Research and Documentation. URL: https://ecprd.secure.europarl.europa.eu/
Eras, H.-J. (2025). Taking Stock of Artificial Intelligence from the Perspective of Parliamentary Reporting in 2025. – Tiro 1/2025. URL: https://tiro.intersteno.org/2025/06/taking-stock-of-artificial-intelligence-from-the-perspective-of-parliamentary-reporting-in-2025/
Fore, P. (2025). Microsoft researchers have revealed the 40 jobs most exposed to AI – and even teachers make the list. Fortune. URL: Microsoft researchers have revealed the 40 jobs most exposed to AI—and even teachers make the list | Fortune
IPU Centre for Innovation in Parliament (2025). Towards AI Maturity. Meeting summary. Parliamentary Data Science Hub, In-Person Meeting, June 4-6, 2025, The Hague, Netherlands.
Koopman, R. (2025), Waarom zoveel AI-projecten mislukken (en hoe je dat van jou kan laten slagen). DataNews. URL: Waarom zoveel AI-projecten mislukken (en hoe je dat van jou kan laten slagen)
MIT NANDA (2025). State of AI in Business 2025. URL: v0.1_State_of_AI_in_Business_2025_Report.pdf
Myers, S. & S. Thompson (2025). A.I. Is Starting to Wear Down Democracy. New York Times. URL: A.I. Is Starting to Wear Down Democracy – The New York Times
Stacey, K. (2025). Technology secretary Peter Kyle asks ChatGPT for science and media advice. Guardian. URL: Technology secretary Peter Kyle asks ChatGPT for science and media advice | Peter Kyle | The Guardian
Howarth, J. (2025). 60+ Stats On AI Replacing Jobs. Exploding Topics. URL: 60+ Stats On AI Replacing Jobs (2025)
Ryseff, J., B. De Bruhl & S. Newberry (2024). The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed. RAND. URL: The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed: Avoiding the Anti-Patterns of AI | RAND
