An app for spotting bogus documents
Researchers at the ETH Zurich Centre for Cyber Trust have developed an app for authenticating official paper documents. Results from a pilot project in Zurich show that, thanks to the new app, there are hardly any more problems with falsified excerpts from the city’s debt enforcement register.
Finding a flat in Zurich is no easy undertaking. Depending on address and size, the waiting list for a flat can be quite long. And because applicants who are unable to present a clean record at the debt enforcement office are immediately out of the running, the temptation to falsify an excerpt can be great for people who have been implicated in debt enforcement proceedings.
Consequently, proprietors have a strong interest in ensuring that excerpts from the debt enforcement register are authentic. Up until April 2024, they had to submit a request to the city’s debt enforcement offices to know whether or not this was the case. Yves de Mestral, president of the City of Zurich association for the heads of the municipal debt enforcement offices, says, “We used to receive twenty to thirty enquiries about falsified documents every day.” He adds that forgeries were detected or reported once or twice a week—and that the number of undetected cases is likely high.
Now, however, help is given by an app developed by researchers in David Basin’s group. Basin is head of the ETH Zurich Institute of Information Security and co-leader of the Centre for Cyber Trust, which receives funding from the Werner Siemens Foundation. He and his team have developed a method to make it easy to authenticate paper documents using a digital tool. To detect forgeries, their app “thenti” compares a copy of a document with the original.
Falsified excerpts all but eliminated
The pilot project in Zurich works as follows: since April of last year, all excerpts from the city’s debt enforcement register have been issued with a QR code. When proprietors or lessors scan the excerpt with thenti, the app analyses whether the information matches the digital copy of the original that has been encrypted and stored on a server. It flags any discrepancies found, and users can then assess these differences to decide whether the document is trustworthy or whether it’s fraudulent.
The project has been an absolute success. “Enquiries about falsified documents are down significantly, at the most we get five to ten per day,” Yves de Mestral says. “It makes the work processes for our staff much easier, although it’s difficult to quantify the difference in terms of costs.” He says he sees practically no more falsified excerpts from the debt enforcement register. “Our assumption is that the QR code is a strong deterrent.” Frank Loeliger, service manager of e-government at the City of Zurich Department of Organisation and IT, says that, from a technical point of view, the app delivers what it promises: “It works without a hitch, we’ve had no failures.”
The researchers, too, give their product good marks. “It’s an exciting project especially because, as far as we know, this is the first time in the world that a digital system has been used to test the authenticity of paper documents in a real-world, uncontrolled setting,” David Basin says. This makes the app’s smooth roll-out all the more satisfying, adds Jesus Solano, PhD student in Basin’s group and co-founder and managing director of ETH spin-off thenti. He also says that app users have only very rarely requested support: “That means the technology and the algorithms we developed are robust and comprehensible.”
Strict data protection
In total, the City of Zurich debt enforcement office issues roughly 2600 excerpts from the debt enforcement register every week. The researchers say that roughly 300 documents per month are currently analysed using the app. Some 300 people actively use the app, which may not sound like much. However, the excerpts authenticated using thenti should be understood in relation to the number of rental contracts concluded every month rather than in ratio to the number of excerpts issued. This is because only excerpts from applicants who are actually selected are likely to be checked. Solano estimates that most app users are professional property managers or lessors who check many more than just one excerpt.
Neither the researchers nor the city officials can say how many forgeries thenti has detected because the app is designed to ensure that all data remain with the app user—in line with data protection standards. “When a falsified excerpt is discovered, the police will probably be informed. We’re not involved,” Jesus Solano explains.
Despite the app’s success, Solano sees potential for improvement, in particular in terms of detection accuracy. The researchers have observed that some users (probably unintentionally) make life a little difficult for the app: they scan documents in uneven lighting conditions, or they fold the paper so that the app reads a straight line as a curve. While the human brain can easily compensate these minor changes, the app’s algorithm is challenged. “We’re now working to develop neural networks capable of dealing with these kinds of situations,” Jesus Solano says.
Other applications
The researchers also want to make the app even more user friendly. Currently, WSS project partner Matthew Smith in Bonn is conducting a study with several dozen participants to discover how well the app aids users in assessing a document’s authenticity and detecting falsifications that range from the obvious to the highly sophisticated. “In addition to finding out how good the algorithms are,” Solano says, “we also want to learn how user-friendly the app is.”
Everyone involved in the project agrees that digitised authentication of official paper documents has potential. Frank Loeliger believes a roll-out of the app in other communes and cantons would be a good idea, while Yves de Mestral says it would be interesting to conduct a study to assess whether these solutions could be used for authenticating other official documents.
While the details have yet to be settled, one thing is certain: the new software would be suitable for analysing many other documents. Examples range from civil status records, birth and death certificates on to professional diplomas and bank statements. And in future, these methods could even be used to detect forgeries of three-dimensional objects like valuable watches or luxury handbags. David Basin says such new applications have met with interest: “We’ve already received additional requests, but it’s too early to say anything specific.”
