Deepfakes Force Polish Investigative Journalists to Change Methods

A few years ago, an authentic recording or photo was, for an investigative journalist, damning evidence. Today it's only the beginning of the work - the material has to be verified, because tools like ElevenLabs can fake someone's voice convincingly enough that telling a fake from the original borders on the impossible. Wirtualne Media described, in a piece published on July 30, 2026, how Polish investigative reporters and lawyers are coping with this new reality.

The End of Undisputed Evidence

Szymon Jadczak, a journalist at Wirtualna Polska, admits outright that today he would not publish a recording without thorough technical verification. That's a shift from the practice of just a few years ago, when raw audio or video material was treated as sufficient evidence for publication. Now newsrooms have to assume that any recording could have been fabricated before they even begin working on it.

If I received a recording like that today, I wouldn't dare publish it without very thorough verification - Szymon Jadczak, journalist, Wirtualna Polska

Attorney Krzysztof Czyżewski of the Czyżewscy law firm points out that the problem runs deeper than a single fake recording. In his view, no document, photo, or recording can be trusted unconditionally anymore, which forces a redefinition of the legal concept of special journalistic diligence - it can no longer be treated theoretically or formally, but requires a real, documented verification procedure.

You can no longer trust any document, photo, or recording unconditionally - Krzysztof Czyżewski, attorney, Czyżewscy law firm

Detection Tools Get It Wrong Too

The problem doesn't end with forgeries. Mateusz Cholewa, author of the book Fake Off, warns that tools for detecting AI-generated material have their limits and have repeatedly flagged fully authentic material as artificially generated. That creates a double risk for newsrooms - they can both let a fake through and wrongly discredit genuine material.

Anna Gielewska, editor-in-chief of Frontstory, points to an even harder-to-detect threat. In her view, the hardest material to catch isn't content generated by AI from scratch, but authentic photos or recordings that have only been partially manipulated. Such hybrids slip past many automated detectors because most of the content in them is genuine.

Specific Cases, Specific Consequences

The article cites several examples in which AI played a role on both sides of an investigation. In the case of fake documents used by Telewizja Republika, it was AI-assisted analysis that uncovered irregularities in VAT rates, exposing the forgery. After the release of files in the Jeffrey Epstein case, meanwhile, a wave of AI-generated material appeared trying to pass as authentic documents from the case - evidence that today, nearly every high-profile source publication almost automatically triggers an accompanying wave of fakes.

Radosław Gruca describes a case in which stylometric analysis confirmed the authenticity of prosecutorial transcripts, but prosecutors ignored it anyway. That points to a competency gap in law enforcement agencies, which aren't always equipped to assess digital evidence verified through computational linguistics or technical analysis.

Media outlets are under growing pressure to produce as much material as possible, at the lowest cost and in the shortest time possible - Radosław Gruca, investigative journalist

Time Pressure Versus Accuracy

That statement from Gruca captures the tension at the heart of the article. Verifying material for possible AI manipulation takes time, specialized knowledge, and often the involvement of technologists from outside the newsroom - resources most Polish media outlets don't have to spare. At the same time, the pressure to publish quickly, sharpened by competition online, pulls in exactly the opposite direction.

Attorney Czyżewski adds another factor - in his assessment, the court of public opinion reacts faster and more mercilessly than the court of law. A newsroom's mistake, whether publishing unverified material or wrongly discrediting genuine evidence, spreads on social media before a case ever reaches a courtroom. That raises the stakes of every publication decision even further.

For Polish readers and for newsrooms themselves, the article's conclusions are practical. There's a growing need for media outlets to build technological expertise - hiring specialists in detecting manipulation, running a documented verification procedure, and treating every recording or photo as material that requires checking rather than ready-made evidence. Sources in the article agree that Polish courts and law enforcement still haven't kept pace with this rate of technological change, which in practice shifts the burden of verification onto newsrooms themselves.