re:publica26 team outing

This year, re:publica felt like an intense stocktaking of our digital civilization, wavering between technological self-empowerment and the defense of democratic foundations. My review inevitably begins with Angela Merkel, who, as a “political citizen” at the WDR Europe Forum, radiated an almost prophetic calm. She reflected very openly on how the internet has changed political work. In view of the current upheavals, she soberly observed: “I believe that digitalization, through the many so-called social media, has even made it harder to have stability”.

Asked about a possible role as a mediator in the war against Ukraine, the former chancellor gave a clear no, as she is convinced that such mediation is inseparably tied to current political power and therefore lies with those in office today. “Personally, it wouldn’t have occurred to me to ask a mediator to go to Minsk for me and talk to Putin like that. […] These are politicians acting today.” Her plea for an inclusive notion of “the people” was also powerful: “The AfD is trying to define who the people are. And I can only ask everyone in this hall and everyone watching: every German citizen is the people. And nobody gets to divide that up. You are the people and you are the elite. I am the people and you are all the people and together we are the people. And these are the things: compromise, facts – and every German citizen has the same rights and duties and belongs to the German people.” When she conceded that it was important, above all, not to keep defining oneself indirectly through the AfD, there was a lot of applause. She warned emphatically that moods must never replace facts, otherwise we are “really on the wrong track”.

Carline Mohr summed up in “Of course nobody’s listening to you!” that many politicians and CEOs have long since lost touch with this fact-check and with today’s reality. She diagnosed a deep crisis among power players who still communicate as if it were 1987, while the world around them has long since gone digital. In this context, it became clear that the technological upgrade through AI often only makes the problem worse, since AI-generated texts usually remain alarmingly mediocre and merely flood the digital space as “AI slop”. This mediocrity means that real attention gets lost in a sea of generated content – and in the end, nobody is listening after all.

A technological counter-model to this powerlessness of the “old world” came from the talk “OpenClaw – Anatomy of the Coming Wave” by Stephan Noller and Benedikt Köhler, which, incidentally, had what were for me the best slides and the sleekest slide design of the entire conference. Their solution “HybridClaw” shows that powerful AI agents don’t need “a gigawatt data center” but can be run sovereignly on a Raspberry Pi in your own network. It immediately made me want to self-host LLMs from now on instead of turning every thought into expensive tokens that only fill the coffers of the big AI tech companies. Projects like OpenClaw or HybridClaw are essentially “friction in practice” instead of a Silicon Valley vacuum, since you can define and trace every step of the machine in simple text files. This technical independence only works, however, if we truly master the possibilities on offer – something Hendrik Heuer, in his talk on “vibe coding”, put under critical scrutiny. English may now be the “hottest new programming language”, but Heuer also warned against the fallacy of speed. The problem is that with “vibe coding” – the rapid generation of code by AI – the most important groundwork, the clarification and testing, is often simply skipped at the level of detail it requires. Software is behavior, and that behavior has to be safe in complex situations instead of going online untested, where it can endanger real people and organizations.

This loss of control and understanding at the technical level is mirrored in political dependence, which Annette Zimmermann, like many others, called the “broligarchy”. She also criticized politics’ “learned helplessness” toward tech giants and demanded that the power to shape AI be returned to the citizenry as a whole (the demos). Just how manipulable these systems are was shown by the “AI stress test” on the topic of milk, in which Simone K. Frey and Jakob Vicari demonstrated once again how blatantly AIs tend toward “sycophancy” (brown-nosing) and often just tell us what we want to hear instead of delivering facts. Global models fail especially when it comes to regional knowledge: one AI wrongly claimed that under Germany’s “housing level 1” label, cows “are permanently kept in the barn and often tethered” – a piece of information that is simply wrong.

Just how creatively – and revealingly – today’s technical possibilities can be used was demonstrated by the investigative session “Talk with Elon”. Tech journalist Eva Wolfangel reported how, in March 2025 – after initially ignoring a message on Signal – she decided to play along with an “Elon Musk” scammer. She posed as a journalist with marketing expertise and actually managed to get “hired” by the scammer for SpaceX and Tesla. In doing so, she turned the tables and used AI heavily against the fraudster herself, having ChatGPT create highly professional marketing concepts and presentations to keep the scammer happy. At the same time, she deliberately built “friction” into the process by inventing, with ChatGPT’s help, two fictitious tax advisors – Dr. Hartwick and Eduard Gundlach – who made the scammer’s life difficult and kept demanding new documents.

The climax of this “counterattack” was an almost absurd form of time theft: Eva Wolfangel sent the scammer AI-generated images of Steam gift cards. Since the codes visible in the images didn’t work, she forced the fraudster to painstakingly type the numbers in by hand and get annoyed by error messages. To confuse the scammer completely, she finally sent him an AI-generated photo of the “real” Elon Musk holding exactly these fake cards, with the cheeky remark that she had already sent the cards to the “real” Elon. What developed was a bizarre “postmodern endgame” in which a scammer bot and a journalist faced off until the fraudster asked, exasperated: “What the fuck are you the real scammer?”. With this experiment, Eva Wolfangel showed that AI isn’t just a tool for criminals but, with humor and creativity, can also be used for education and defense.

But while we’re still laughing about scammers, the state is arming itself with “Palantir laws” that deeply infringe on our fundamental rights. In the session by Franziska Görlitz and Constanze Kurz it became clear that we are at a critical tipping point. The US company Palantir, a longtime military contractor, provides software with its product “Gotham” (in Germany often known under names like HessenDATA) that the company itself advertises as a “digital weapon”. Behind the company stand figures as dazzling as they are controversial, such as billionaire Peter Thiel (an early Trump supporter) and CEO Alex Karp, who cultivates a “mad scientist” image and proudly positions his company as completely “anti-woke”. In the US, the software is already being used to locate and deport people – images that should serve as a warning of what is technically possible with such tools.

In Germany, the software is in use in various federal states. The problem lies in the legal basis for these automated data analyses, which allow the police to merge huge, heterogeneous data sets in order to generate “new knowledge”. In the process, data from completely law-abiding citizens is analyzed en masse – people who ended up in police databases as witnesses to an accident or victims of bicycle theft, for example. This leads to a massive “chilling effect”: people start changing their behavior and avoid demonstrations or political meetings just so they don’t mistakenly end up in an algorithm’s crosshairs. This also affects, for example, (mediating) fan liaison officers in soccer, who are naturally always found near violent fans in order to assess and defuse situations. Although in a landmark 2023 ruling, the Federal Constitutional Court decided that such analyses are only permissible in the case of specific threats to particularly important legal interests and that lawmakers must give the police strict guidance, this is often ignored in practice. In Hesse, for example, the legal “guidance” on avoiding algorithmic discrimination consists of just a single, vague sentence, which experts criticize as a blatant disregard of the ruling. Moreover, there are still no independent scientific evaluations to prove that these multimillion-euro tools actually make threat prevention more effective. Instead, the federal states are entering into a dangerous dependency on a private US company (vendor lock-in) that gains access to sensitive police insider knowledge. Particularly worrying is the trend to extend these powers to intelligence services such as the Federal Office for the Protection of the Constitution, often even without a sufficient legal basis. If we don’t stop this development, our democratic civil liberties risk being gradually eroded by algorithmic mass surveillance.

One of the most sobering points of the conference overall was the engagement with digital fascism. It is alarming how the AfD uses AI-generated content and conventionally attractive avatars like “Larissa Wagner” to build parasocial relationships with voters and spread disinformation. These strategies were examined in several talks, along with warnings about an “authoritarian overhaul” that often begins unspectacularly in the “dark corners” of ministries and can paralyze critical civil society by cutting budget items. Max (Maximal Demokratisch) spoke about the personal dangers and threats, the kind you face when you stand up to the right-wing mainstream in small towns. He called on journalists and creators to leave “disaster mode” and to also share positive democratic news from regions like Saxony-Anhalt, so that right-wing dominance doesn’t become the norm. In this context, Arne Semsrott also called for building “counterpower”, in order to counter the authoritarian turn with structures of our own and productive conflict.

re:publica 26 was thus once again a loud wake-up call. It is our duty to actively confront algorithmically driven hate and to regain digital sovereignty before the line between fact and fiction is torn down for good.