AI-Powered Cybercriminals Stole 537 Million Zloty from Poles in 2025

Cybercriminals are increasingly turning to artificial intelligence to drain Poles' savings. Poland's Central Bureau for Combating Cybercrime (CBZC) closed out the year with numbers that show the scale of the problem: losses in cases handled by the Bureau reached 537 million zloty in 2025, up from 115 million zloty a year earlier.

The figures come from CBZC's official year-end summary and a report by CERT Polska (Poland's computer emergency response team), which handles security incidents on the Polish internet. Both documents paint the same picture: the number of fraud attempts is growing much faster than law enforcement's capacity to pursue them, and losses are climbing as generative AI techniques become more convincing.

The scale of the losses

The jump in losses from 115 to 537 million zloty between 2024 and 2025 amounts to nearly a fivefold increase in a single year. CBZC notes that this covers only cases the Bureau itself handled, so the real scale of losses among all Poles is likely higher, since some victims never report the fraud to police at all.

Last year officers seized more than 80 million zloty from suspects, of which nearly 25 million zloty was recovered and returned to victims. That is still a small fraction of the total losses, showing how hard it is to recover money transferred to accounts controlled by organized crime groups, often operating from abroad.

How AI powers the fraud

The most lucrative scheme remains fake investment platforms promoted using deepfakes of well-known people. In one case broken up by CBZC, a criminal group operating under names such as Agricole Trade, Partner Groupe and Global Maxis used this method to swindle around 75 million zloty from more than 1,500 people who had seen online ads featuring the faces of politicians, businessmen or celebrities supposedly endorsing the investment.

Alongside video deepfakes, voice cloning is showing up more and more, used in so-called impersonation scams, posing as a police officer, a bank employee or a family member. Losses from this type of attack reach around 140 million zloty a year in Poland. Victims who hear a familiar voice on the line grow suspicious far less often than they would with a standard call from an unknown number.

We assess the Bureau's current level of development at 60 percent, we are ready for new challenges - Central Bureau for Combating Cybercrime (CBZC)

CERT Polska's numbers

CERT Polska, the team that monitors threats on the Polish internet, logged 658,300 cybersecurity reports in 2025, of which nearly 260,800 were confirmed as actual incidents. That is a 152 percent increase compared with the previous year, driven partly by AI-generated phishing campaigns that are harder to spot than earlier versions riddled with language errors and unnatural grammar.

Fraudulent transactions in electronic banking form a separate category of losses. According to data from the Narodowy Bank Polski (NBP, Poland's central bank), 800 million zloty vanished into fraudulent transactions in the Polish banking sector last year. Most of these operations do not stem from break-ins to banking systems, but from customers themselves, persuaded through social engineering, authorizing the transfers.

The American comparison

The scale of the problem is visible across the Atlantic too. The FBI's Internet Crime Complaint Center, IC3, calculated for the first time the losses stemming specifically from AI-enabled fraud: 893 million dollars from more than 22,000 reports last year. Total losses for Americans from internet fraud reached about 21 billion dollars, with people over 60 hit hardest, reporting 7.7 billion dollars in losses.

Criminals use pressure tactics to deceive victims through fake social media profiles, cloned voices, forged identity documents and convincing videos featuring public figures or family members - FBI

What this means for Poles

For the average internet user, the practical takeaway is simple: spotting fraud by accent, language errors or poor recording quality no longer works. Experts quoted in Polish media point out that awareness of the term deepfake itself is still limited in Poland, especially among older employees, who do not always know that a realistic recording can be made using a boss's or a family member's face and voice.

In its summary, CBZC forecasts that 2026 will bring even wider use of AI tools by criminals, and with it, fake content that grows ever harder to recognize. Financial institutions and law enforcement will have to keep pace with technology that can now generate a convincing video of any public figure in a matter of minutes.