Artificial Intelligence [AI] : Protecting online dating users from fraud
Cyber crime and dating fraud is a growing business for criminals. Hundreds of millions of dollars is stolen each year.
|USA||$219 million 2017||Source FBI|
|UK||£27 million 2016||Source Action Fraud|
|Australia||$42 million 2016||Source Australian Competition and Consumer Commission|
At saucydates.com we have worked for 10 years to tackle the criminals behind dating scams. Scarlet is our latest development to educate and protect users. Scarlet is an artificial intelligence virtual assistant that interacts with our users when they are having online conversations with others. She offers advice as well as triggers manual profile previews for our moderation team. The advice she instills is only viewable to the potential victim, a fraudster is unaware that Scarlet is reading, learning and advising.
The example conversation above is common for scammers, they want users to leave a site as soon as possible so they cannot be detected and removed. Additionally, once they have a users email it’s added to a ‘suckers’ list and exposed to more scams and phishing attacks.
How does Scarlet work?
We have two traditional ways to detect fraud which are common place within the dating industry. The first is our moderation team, who can flag a suspicious user. The second is our members, who can report suspicious activity; member reports are broken down into different categories such as abuse, fraud, underaged user etc. A human moderator then checks the activity of a reported user and decides if the account should be terminated.
This base line enables us to split messages into different sections and within the fraud section we can identify different types of scam. Scarlet is then coded to detect and learn within each type of sub fraud. Some examples types she knows are:
Traditional romance scam:
The victim falls in love with a fake person who will eventually need some financial help which is a scam.
The victim is asked to get a local hotel to meet up for a night of passion and asked to share the hotel name and reservation number. The scammer calls the victim, pretending to be the hotel, explaining there’s a problem with the payment and do they have another card or want to try the payment again. The victim as now given their credit card details to a scammer.
The scammer pretends to be a webcam model and asks men to join her (his) site to watch. The scammer gets paid a commission by the cam site for each new member. Cam site owners know this happens but some can be lax in addressing it as they are making money from it too.
The scammer pretends to be a glamorous female and gets men to perform naked sex acts on a webcam. They record the camera session and then blackmail the user by threatening to share it with the victims LinkedIn ad Facebook connections. Tools such as Google image search allow a criminal to locate a social network account from a users dating profile picture.
Gift card scam:
A scammer claims their smart phone’s apps are out of date which they cannot afford to update. They request an app store gift card to enable them to communicate further.
Breaking down the problem into subsets makes it quick for Scarlet to become an expert. Taking 100 messages that have been identified by moderation as a hotel scam she can find patterns and similarities.
Presently Scarlet will only flag a user for manual moderation and will not delete them automatically. But she helps with the moderation by re-reading all the messages and colour coding them for the reviewer. If a human reviewer is presented with a screen of red messages then it makes the decision to ban the account quick and easy.
Notice in the screenshot above of our moderation control panel that Scarlet has found all messages related to leaving the dating site but ignored general messages.
Making our users aware of scams is imperative to our company, many brands hide the problem which is rife within the industry. We feel education not only protects our users but helps them spot and report more issues, which is turn helps Scarlet improve her knowledge of scams.
The virtual assist aspect of Scarlet joining in with conversations has been a major break through in fraud prevention for our dating network. If you have static fraud advice on a single page you need someone to read it, remember it and apply it. It’s a big ask when a user is taking to someone they are attracted too. Having Scarlet within the live conversations is a fundamental change for the better.
Since implementing Scarlet a few months ago the number of fraudulent users has dropped by 75%. Unfortunately, the criminals still exist so will move to another dating sites with weaker tools or a disregard for fraud awareness.
Reducing online dating fraud
The media and public’s views on technology and privacy differs considerably but to reduce online dating fraud these are the areas that we feel still need to be addressed:
- Dating site owners need to educate their users and innovate to help reduce the effectiveness of scammers via moderation and AI.
- End to end message encryption makes moderation and AI near impossible and should not be implemented for dating sites.
- Proxy networks and VPNs that allow anonymous browsing and effectively change the IP address location of users need to be disallowed for dating sites. A public database of sites opting out of proxy networks would help significantly, making a network owner responsible for blocking access.
- Mobile handset manufactures should make it impossible, or at least very difficult to fake a GPS location with a third party add on.
Often there is a misunderstand that IP addresses are the key to fraud prevention but VPNs, proxy servers, mobile networks, wi-fi and dynamic broadband addresses mean the ownership on an IP address is either shared or temporary. The biggest single impact to fraud reduction would be for every device to have an unique ID that is publicly visible but online privacy would then be abolished. Today, it’s creative thinking and AI that are the best defence in online dating fraud.