Germany’s AI ‘Insult Detector’ Scans X for Lawsuits—but Did Taxpayers Really Give SoDone €1 Million?
SoDone uses AI and legal review to find potentially unlawful posts for politicians and other clients. Its business model is real; viral claims about nearly €1 million in state aid are not supported by the records found.
A German legal-technology company called SoDone uses artificial intelligence to scan social media for potentially unlawful insults, threats and false factual claims. Politicians across major parties have used the service, prompting a provocative question: are algorithms protecting victims from mass abuse, or industrializing lawsuits against political speech?
The core business is real. Tagesschau reported that SoDone's clients have included prominent figures such as Robert Habeck, Marie-Agnes Strack-Zimmermann, Ralf Stegner, Hendrik Wüst and Roderich Kiesewetter. In one batch, the company sent around 400 complaints for Kiesewetter to Baden-Württemberg's criminal police, creating a significant investigative workload.
The AI does not itself convict anyone. It filters large volumes of posts for language that may be criminally relevant. Staff and legal partners review the results, clients decide which matters to pursue, and police, prosecutors or civil courts determine what crosses the legal line.
SoDone also finances litigation. Public descriptions say victims use the service without paying upfront, while the company assumes costs and receives 50% of successful compensation. If a defendant cannot pay or a case fails, SoDone bears business risk. That differs from a government automatically fining users because an algorithm dislikes a tweet.
Critics still identify serious dangers. Automation can make hundreds of marginal cases economical, overwhelming police and pressuring ordinary users to settle rather than pay for counsel. German law criminalizes some personal insults more broadly than U.S. free-speech doctrine, and protections for politicians have generated controversy when harsh satire or anger becomes a police matter.
The viral funding claim is much weaker. Public records show SoDone won third place in the 2024 MUT startup competition in North Rhine-Westphalia, receiving €10,000. A contemporary Welt report described the private company as receiving no public subsidy and not seeking one. No credible record located supports the claim that taxpayers gave SoDone nearly €1 million.
The figure may confuse SoDone with HateAid, a separate nonprofit that has received substantial public and private funding, or with broader government programs addressing online hate. Combining organizations with similar missions creates a dramatic narrative but an inaccurate balance sheet.
A second viral claim says SoDone paid €11,000 after politicians dropped a case challenged by lawyer Markus Haintz. Lawyers publicly post examples of withdrawn complaints, but the specific payment, legal basis and responsible entity require court or settlement documents. A dropped criminal charge, an abandoned civil demand and an award of defense costs are different events.
Supporters of the service point to threats, doxxing and repeated sexualized abuse that drive people—especially women and minorities—out of public life. Telling targets to ignore thousands of messages effectively grants impunity through volume. Technology can preserve evidence and make existing rights usable.
Opponents answer that powerful clients already possess visibility, institutional support and legal access. An automated pipeline can chill criticism when citizens cannot predict whether a crude label will be dismissed as protected opinion or produce a costly letter.
Germany's Basic Law protects expression but allows limits through general laws and personal honor. The democratic challenge is maintaining a clear boundary between true threats and unlawful defamation on one side, and offensive political judgment on the other. Human review helps, but incentives matter: a company paid from successful claims may favor aggressive interpretation.
The criminal and civil routes should not be conflated. A criminal complaint asks the state to investigate and can involve police identification of an account holder. A civil warning letter can demand deletion, an undertaking and money. Prosecutors may close a criminal file while a civil claim remains possible, or a court may reject a payment while still finding language abusive. Viral summaries often compress these outcomes into “the politician lost.”
Algorithmic accountability would improve the debate. SoDone need not reveal code or private client material, but it could publish precision rates, appeal outcomes and demographic or political bias testing. If the system disproportionately flags insults against officials while overlooking abuse directed at ordinary users, its “digital shield” would look less like neutral access to justice.
Public authorities also need triage rules. A flood of automatically prepared files may be legally valid yet divert investigators from credible death threats, stalking and coordinated harassment. Scale is beneficial only if it improves prioritization rather than replacing it.
What to watch next
Watch transparent figures for posts scanned, cases recommended, complaints dismissed, civil judgments won and public money received. Courts should publish enough reasoning to show where criticism becomes illegality. Can SoDone demonstrate that it targets genuine abuse with low false-positive rates, or will scale transform a legitimate remedy into a private enforcement market that makes Germans fear criticizing those who govern them?