Deepfakes about pensions are reaching a large audience. BR24 found nine YouTube channels and eleven TikTok channels with hundreds of AI videos on pension-related topics. A single clip received almost 140,000 views. At the same time, Google Trends in Germany recorded more than 100,000 searches for the term “pensioner” and an increase of approximately 1,000 percent.
The videos combine professionally-looking AI avatars with false numbers and urgent messages. The case concerns Germany, but the transparency rules of the European AI Act are also relevant for people in Austria.
What are the claims made by deepfakes about pensions?
iksHaus | KI-generiert nach redaktioneller Konzeption, 2026
In a video examined by BR24, an older-looking man in a suit appears as a supposed pension expert. He claims that 2.2 million pension statements were incorrect due to calculation errors. The German Pension Insurance Fund and the social association VdK contradicted this statement.
According to the approximate figures provided by the German Pension Insurance Fund to BR24, around two million pension statements were issued in 2025. Approximately 167,000 of these were subject to an objection process. In reality, errors occurred in 1,400 cases. While individual errors are not excluded, the figures contradict the viral claim of millions.
Based on the approximate initial values, the error rate is approximately 0.07 percent based on two million statements and approximately 0.84 percent based on 167,000 objection processes. These values cannot be applied to all pension statements. However, they show the scale of the issue that the video is concealing.
Other investigated clips claimed that the state had stolen 240 billion euros from the pension fund, or that a ruling by the Federal Constitutional Court would overturn the entire pension policy. These statements also did not withstand fact-checking. The German Pension Insurance (Deutsche Rentenversicherung) stated to BR24: “Often, the focus is on financial implications that do not actually occur in reality.”
Why do artificial experts seem credible?
iksHaus | KI-generiert nach redaktioneller Konzeption, 2026
Maria Pawelec, a deepfake researcher interviewed by BR24, points to the negativity bias. Fear, anger, and the prospect of a loss attract particular attention. In the context of pensions, reports about money, security, and potential errors can therefore create a high sense of urgency.
In addition, there is the confirmation bias. Those who already distrust institutions may more easily accept a claim that confirms their suspicions. This is not a characteristic of a specific generation, but a general human pattern. AI producers can deliberately reinforce this by using calm voices, direct address, suits, and bookshelves. These features convey competence, but do not prove qualification or editorial responsibility.
Richard Kuchta from the Institute for Strategic Dialogue told BR24 that there is no data to suggest that older age groups are more likely to believe deepfake videos. In the ARD-ZDF media study 2025, 2.25 percent of people over 50 reported using TikTok the previous day. For YouTube, it was 11.1 percent. The clips can also be forwarded via family members, search results, chats, and recommendations.
BR24 found links to a fraudulent website for dietary supplements and medical products on one of the YouTube channels investigated. No comparable external fraudulent links were found on the other 19 accounts. Therefore, monetization or political motives are not proven for every channel. It is certain that comments and viewing time can increase reach.
What changes will the AI Act and platform regulations bring?
iksHaus | KI-generiert nach redaktioneller Konzeption, 2026
Since August 2, 2026, the transparency obligations outlined in Article 50 of the AI Act generally apply throughout the European Union. Deepfakes must be labeled. Furthermore, providers of generative systems should include machine-readable markings for artificially generated or manipulated content. For certain older systems, a limited transitional period applies until December 2, 2026, regarding technical requirements.
According to the European Commission, violations of Article 50 can be penalized with fines of up to 15 million euros or up to three percent of the company’s global annual turnover. The Commission summarizes the goal as follows: “People should know when they are interacting with AI or exposed to AI-generated content.”
BR24 found that most of the YouTube videos examined were labeled as AI-generated. Many of the corresponding TikTok videos lacked such labeling. TikTok informed BR24 that reported videos were removed due to violations of the platform’s rules. However, similar accounts remained accessible. According to its spam policy, YouTube prohibits the automated production of large quantities of very similar content with only minor changes.
A label alone does not determine the veracity of the content. A context check is more informative: Who operates the account? Is there an “Imprint” (legal notice) and a traceable institution? Do the dates and figures mentioned correspond to information provided by the German Pension Insurance (Deutsche Rentenversicherung), legal sources, or established media outlets? The presence of many nearly identical clips featuring the same figure is another warning sign.
In May 2026, the German Pension Insurance also warned against deceptively realistic phishing emails that imitate AI-generated language and phrasing. According to the Pension Insurance, it never requests personal data via email. Therefore, source verification can protect against both misinformation and fraudulent attempts.