How Exposed Employee Data Fuels AI-Powered Social Engineering

In September, European fintech giant Revolut handed passport images, selfies, and full transaction histories to someone posing as a government agency. Nobody broke in or hacked anything. The request came from a real government email domain, it had valid credentials and so Revolut complied with the request.

That’s how a social engineering attack ends. Here’s what it looks like from the very start. 

A team of European researchers collected the work email addresses of 7,741 university employees and ran each one through a search engine. About half kicked back some results. A large language model was supplied with those results and a prompt: Write a phishing email for each of those individuals. The cost: less than three cents per email. Those personalized messages were clicked nearly three times as often as a generic email. 

What about those 3,731 individuals whose search queries came back bupkis? With nothing to personalize, they got the generic email. They clicked less than half the rate of the cohort that Googled well. 

Social Engineering Starts with Reconnaissance

Most companies train their employees to spot the attack when it arrives in their inbox or via text on their work device and even via real-time voice. How well can the average employee (is there such a thing?) recognize that an email, text or phone call is suspicious? Training helps. But these days, by the time the threat actor actually dials a number or hits send on some other form of communication, the target is toast because the attacker knows exactly who they are contacting, what a normal request looks like, who their target reports to, and who needs to sign off on whatever it is they are requesting. 

There is no force. There is making moves that disappear into the flow of everyday life at the target organization.

Revolut wasn’t an isolated case. That same week, the hardware wallet maker Trezor disclosed that a breach at a third-party email provider allowed phishing emails to be sent from its domain. The attacker didn’t hack either target. They borrowed someone’s credibility. They knew how to talk to someone. Or, more likely, they used AI tools that told them what to say, when and how to say it to get the desired result. No shapeshifting occurred, just garden-variety social engineering. 

How Data Brokers Turn Employee PII into Pretext

Consider how quickly you can put together a dossier on someone you’re thinking about doing business with. Threat actors do the same with the breadcrumbs of identity we allow to live online. It could be a dinner attended, a LinkedIn post, a tagged photo on social media, a road race, a hobby. They also use the data readily available online from brokers. 

Knowing that an employee works in finance is useful. Knowing who they report to, where they used to work and who else works at the company is better. Threat actors are clocking who likes what on LinkedIn. They are going online and getting background on those people. The best ones are not coming for a target till they know about the fishing trips, the golf outings. Add a personal cell, readily available online, and spouse name with your home address—the pretexting can make it very hard to discern the threat actor behind an innocent-seeming communication.

Make no mistake. Much of the information threat actors need is free, and whatever else they might find useful can be bought for a minimal fee. In 2023, researchers at Duke University in a study funded in part by West Point, bought personal data on active-duty service members and veterans from U.S. data brokers for as little as 12 cents a record. Some brokers sold to a buyer using an overseas email domain and a Singapore IP address. If that’s what it costs for soldiers, consider what it costs for your finance team. 

AI Spear Phishing Costs Pennies Now

Targeted attacks have always been the real danger. 

Barracuda Networks found that spear phishing makes up about .1 percent (that’s point one) of all email-based phishing attacks but is linked to 66 percent of data breaches. The reason it stayed rare back in the day was labor. Open source intelligence is real work. It’s time-consuming, and it requires expertise. In the university study, a senior security researcher writing by hand spent about seven minutes per target. The automated system took 45 seconds. The researchers personalized more than 4,000 emails for about $120. 

Now imagine what a threat actor can do with hours and the same tools. 

Personal attention used to be reserved for founders and the C-Suite, and maybe some key players with the keys to the kingdom. That is no longer the case. 

In the study, every recipient was warned that there was an attack coming. The personalized attacks were created on bespoke AI–small, open-source models; not the most powerful AI tools available. And while the handwritten emails still received a 24.2 percent click rate, more than double the automated ones, the point was well made: AI was going to make the pre-existing problem of our data being everywhere a panic-level threat to every walk of life at a company, and that means more breaches. 

Reduce Your Employee Data Attack Surface

Social media is going to be used. Emails are going to be opened. Human nature is a good thing. And while training is fine, and important, it has to be right-sized. 

It is unreasonable to place the burden of vigilance on employees. Unless they were hired to do security, they have other jobs to do. That’s why it’s crucial to reduce what’s exposed online in the first place.

A good place to start is finding out what an attacker can learn about your employees online. Much of that information may already be sitting on data-broker and people-search sites or social media.

The researchers recommended a layered approach to security that includes multi-factor authentication, limits on what any single compromised account can access, training that focuses employees on the dangers of pretexting and hyper-personalized attacks, and limiting searchable details that can be linked to a specific email address. There is no way to zero-out the risk. Removing the raw material of attacks, however, creates friction in a threatscape that values ease.

Start asking this question: “What does an attacker already know about our employees, and how could that information be used against us?” The answer will be found on people-search sites and other data broker sites. That data will be meaningfully amplified by employee social media use. These factors need to be top of mind for everyone in a secure organization.

Exposed employee and executive PII should be treated as part of an organization’s attack surface. If attackers are using that information to target your people, reducing it belongs in the cybersecurity conversation.

 

Join our LinkedIn group Information Security Community!

No posts to display