The evolution of how brand impersonation attacks use social media

A new academic study of more than 1.3 million social media accounts was given recently at this month’s Usenix conference in Philadelphia. The paper, entitled The Imitation Game: Exploring Brand Impersonation Attacks on Social Media Platforms, makes for interesting reading and sadly shows just how well developed this ecosystem is. Ironically, as business brands pay more attention to social media interactions with their customers, they also enable imposters to launch attacks because people now expect companies to interact with social media. This means that there are many scam accounts that impersonate the brands to create confusion. These lure customers into providing private data and can result in stolen funds and further attacks. The research claims to be the first large-scale measurement of the social scamming ecosystem.

The research team, which was composed of academics from Germany and the US as well as from Paypal, identified almost 350,000 usernames performing various typosquatting techniques to impersonate more than 2,800 brands across Twitter (I know it is called something else, don’t remind me), Instagram, YouTube and Telegram.

Typosquatting is using deliberate typos in user and domain names to make it appear that paypel_support is really the people answering your connection problems. It is not a new problem when it comes to domain names, but as I wrote earlier this year for DarkReading, its use is proliferating in a variety of ways. One way that I didn’t mention is how fraudsters are using it across social media networks. Twitter “is the primary platform for brand impersonation attacks, with fraudsters frequently using typosquatting in their usernames. Roughly a third of these deceptive profiles also use official logos to appear more legitimate.”

The team found that brand impersonation involves multiple steps: after setting up a fake profile (oftentimes using the real brand’s logo to lend legitimacy), the fraudsters engage with customers through posts and offer phony incentives such as discount cards, free services and the like. But the attackers then collect sensitive data, including identities, credit card numbers and other details that are used to engage them in other fraudulent activities.

The most commonly targeted brand is Netflix, which is troubling because right now Netflix is sending out numerous legit messages heralding a change in their account pricing. Apple is the second most targeted brand.

The researchers have several suggestions to try to stem the tide, but admit these will be tough to implement. One of them is pretty obvious: in their work with Paypal, they found that many brands haven’t done their homework and failed to use Know Your Customer methods and continually scan for stolen identities, monitoring their brand mentions online or check for fraud card usage. One recommendation is to send out a quick autoresponse to a customer query to try to engage them before the scammer does. Another is for social media platforms to validate a brand when a new account is created, so that the owner of the proposed paypel_support account really is someone@paypal.com and not fakeuser123123@gmail.

Building the world’s largest digital camera

LSST Camera and SLAC Camera Team-5.jpgThe world’s largest digital camera is a 3200 megapixel behemoth that sits on top of a mountain observatory complex in Chile. Ironically, it was created by engineers that in the past have focused on tracking the universe’s smallest subatomic particles. The camera has one acronym (LSST) and goes by two long names — the legacy survey of space and time, or Large-Aperture Synoptic Survey Telescope.
The camera is part of a telescope at the Vera Rubin observatory, named after an American astronomer that studied dark matter. Everything is still under construction and is expected to become operational next year. When it does, it will work in very different ways than its peers. And while the Webb telescope has gotten plenty of press for flying around the sun these past couple of years — and rightly so, I don’t want to diminish its accomplishments — the  Simonyi telescope at the Rubin is an interesting science tool in its own right. And yes, that name is familiar to many of you. Charles Simonyi worked in the early years at Microsoft, and both he and Bill Gates were early donors to the project.
The Rubin project has been long in coming, just like the Webb. In fact, pieces of it were built in different factories and labs around the world. The camera came from California (the Stanford Linear Accelerator team), the mount was from Spain, and Chile put together the buildings housing everything.
First off, if you have in your mind this is a place where astronomers go to peer through the eyepiece of the telescope and stare at the night sky, put that picture aside. This is a digital camera, and it operates hands-off for the most part. Its goal is nothing short of extraordinary. Every three nights, weather permitting, it photographs the entire night sky, moving around in a pre-programmed pattern. Most telescopes of the past were firmly anchored to their mountain top aeries.
In the past, telescopes like Webb and other expensive instruments required scientists to schedule time on them to focus on particular areas of the sky, and then download what was collected. Committees would vet proposals and schedule the sessions accordingly. Having a telescope that sweeps the entire sky — and doing it in such high resolution — means that you can approach observations in a completely different way.
First off, you don’t download anything. Given the size of the datasets, that would take time, even on high speed bandwidth. All the data stays intact, and you run your queries remotely.
This is a massive amount of data — petabytes worth — and it is all uploaded to an open source repository. Anyone can access the information for their research or just for curiosity. I imagine that schools will jump on board using this archive. It might change the way we teach astronomy and it certainly will reach a wider audience.
Also, the science team behind the Rubin is developing software that mimics what the early astronomers did manually, to seek out changes in the observations. Did a planet move in front of its star? Is a black hole forming someplace? I remember as a child reading about Clyde Tombaugh and how he discovered Pluto (poor Pluto, now downgraded to a demi-planet) in 1930 by looking at photographs taken on different nights to find its movement. He used a device called a blink microscope to quickly flip back and forth between the two photographic plates. Now we have open source code to do that tedious task.
This means that discoveries will be made almost every night, because the universe is a busy place. Scientists don’t have to depend on picking the right time and piece of sky real estate to observe a supernova, but can have software seek out the possible event.
Another distinction: unlike the infrared-based Webb, Rubin operates in visible light.
Finally, what I also liked is that the project is the first time a publicly-funded astronomy effort has been named for a woman.

Tech+Main podcast: The changing role of today’s CISOs

I spoke to Shaun St. Hill, host of the Tech&Main podcast, about the latest YL Ventures CISO Circuit Report. They have a very strong advisory panel of security professionals and annually poll them about industry trends, what their biggest organizational challenges are, and how they interact with their management and boards of directors to protect their companies.

You can listen to the 30 min. podcast here.

Red Cross: Air Force Veteran Works with SAF to Help Service Members

When a veteran retires, most don’t think of setting up their homes on a military base, but that is what Jill Eaves and her family did at Missouri’s Fort Leonard Wood. The Army post is home to the Sixth Infantry Division and one of four major training centers. For the past 80 years has seen hundreds of thousands of members of all four branches of the armed forces train for active and reserve duty, including specialized engineering training. Eaves and her husband of 10 years both served in the Air Force, and when the time came for retirement, they decided to move back on a military installation. After all, with more than 63,000 acres, there is plenty of room. “It is a great place to raise my two children, too,” she said.

Here she is fixing a helicoper while deployed in California. You can read my profile of her for the Red Cross here.

How to make AI models more processor efficient

I was amused to read that a mathematical method that I first learned as an undergraduate has been found to help make AI models more processing efficient. The jump is pretty significant, if the theories hold in practice: a drop in 50x power consumption. This translates into huge cost savings: some estimate that the daily electric bill for running ChatGPT v3.5 is $700,000.

The method is called matrix multiplication and you can find a nice explanation here if you really want to learn what it is. MM is at the core of many mathematical models, and while I was in school we didn’t have the kind of computers (or built-in to our digital spreadsheets or in Python code) to make this easier, so we had to do these by hand as we were walking miles uphill to and from school in the snow.

MM dates back to the early 1800s when a French mathematician Jacques Binet figured it out. It became the foundational concept of linear algebra, something taught to math, engineering and science majors early on in their college careers.

The researchers figured out that, with the right custom silicon, they could run a billion-parameter model for about 13 watts. How do you make the connection between the AI models and MM? Well, your models are using words, and each word is represented by some random number, which are then organized into matrices. You do the MM to create phrases and figure out the relationships between adjacent words. Sounds easy, no?

Well, imagine that you have to do these multiplications a gazillion times. That adds up to a lot of processing. The researchers figured out a clever way to reduce the multiplications to simple addition, and then designed a special chipset that was optimized accordingly for these operations.

It is a pretty amazing story, and just shows you the gains that AI is making literally at the speed of light. It also shows you how some foundational math concepts are still valid in the modern era.

What happens when your plane’s GPS doesn’t work

Many of us have a love/hate relationship with our GPS’s. We love the fact that they can tell us when a route is filled with traffic, or a better way to get from Point A to Point B. But we hate it when we are running late and when the GPS route is a convoluted series of seemingly contradictory turns down small side streets, or when we are somewhere where coverage is spotty or blocked.

That is fine when we are in a car, or using transit. But what happens, as I pose in the subject line, when you are flying a plane and its GPS quits working? It sadly is happening with increasing frequency, as the places around the world that are part of conflict zones continues to expand, and because spoofing or blocking GPS signals is one way to prevent military actors from getting precise positions. That link will document exactly what is happening, and you can click on other links at the end of that post to understand the different types of spoofing that have been observed. The number of incidents has risen alarmingly in the past several months, with as many as 1350 daily flights encountering spoofing, but averaging 900.

Now, that may sound like a lot of flights, but when you consider that these days about 100,000 flights fly every day around the world, it is admittedly still a small number. However, the spots where GPS signals are unreliable have expanded to ten distinct areas. Some of these you might suspect, such as around the Middle East and Russia for example.

(I wrote a few years ago about the Russian airlines. This is yet another reason to steer clear of any flights that come near the place.)

But there are several problems behind this data. First, flight crews are not trained to switch off their GPS when spoofing happens. In fact, they run a variety of automated systems that rely on the global GPS network with all sorts of acronyms. Some spoofing hits the autopilots driving the plane, some hit the ground collision radar that prevents planes from hitting the side of a mountain, others hit the transponders that broadcast the plane’s identity. Second, the symptoms aren’t consistent across all these systems: Each system exhibits different behavior when they get spoofed GPS signals. This means that aircrews are losing trust in their instruments, which is not a good thing. Third, the air traffic controllers — particularly the ones handling long-haul transoceanic flights — have to work harder to separate the planes flying a particular route which are in trouble, which means lower passenger capacity and more flight delays. In some cases, the situation happens close to a landing, which means some planes have to “go around” and take time and fuel to attempt a second landing. Finally, there are a bunch of aircraft-related international organizations that work together in a very delicate balance, and spoofing upsets that particular applecart. Imagine the UN, only worse.

GPS spoofing is now being used as a weapon of war, and it is sadly catching on as the investment in small armed drones is small but the damage that they can cause is great if they can rely on precise positioning for their targets. Sadly, it will take some time before the civil aviation industry can retool to work around spoofing in any effective manner.

CSOonline: Port shadowing is yet another VPN weakness ripe for exploit

A new flaw in virtual private networks (VPNs) was reported last week at a security conference. The flaw, discovered by a collection of academic and industry researchers, has to do with a vulnerability in how VPN servers assign TCP/IP communication ports and use this to attack their connection tracking feature. This flaw, called port shadowing, is yet another weakness in VPNs that corporate security managers have to worry about. As you can see from the chart below, it goes to the way modern VPNs are designed and depends on Network Address Translation (NAT) and how the VPN software consumes NAT resources to initiate connection requests, allocates IP addresses, and sets up network routes.

I write about this issue for CSO here.

Having a stranglehold on the world’s economy, thanks to a single bad file

By now you have no doubt read a lot of bad coverage about the Crowdstrike update that crashed more than 8M Windows systems around the world. We have been treated to all sorts of action photos of the “blue screen of death” showing up in airports, across the exterior of the Vegas Sphere, and other places that have interactive displays that aren’t interacting with anything. All because of a bad virus definitions file that was put out there for little more than a hour on Friday. This file was placed in a very sensitive area of the Windows OS (I will get to that in a moment) and because it was poorly written and inadequately tested.

I say virus definitions file because that is really what it is. Crowdstrike released this after-action report that is filled with doublespeak, jargon, and a tremendous lack of clarity earlier today. What is interesting from a corporate comms perspective is that they explain things in detail that we don’t need and ignore things that we really want to know. What they don’t say can be found on Kevin Beaumont’s blog.  Here is what I have gleaned from the sad affair:

  • Almost every anti-malware, endpoint detection and remediation, SASE whatever you wanna call ’em vendor has a similar configuration of their products. They have to have this unfettered access to the Windows kernel because that is how they work, and have worked since the early days of Norton Anti-Virus running on DOS. So just because you use some other vendor’s product doesn’t mean this can’t happen to you.
  • These vendors have to update their software frequently to stay ahead of exploits that they find, and this means that they are sending out stuff fairly frequently.
  • This means that the chances of a badly crafted file can be created in the rush to get these updates out. It was a small miracle that the file was only online for little more than an hour.
  • Beaumont says that we are “handing the keys to the global economy to a small group of private cybersecurity companies with no external governance or assurance. It has always felt sketchy, and today feels very sketchy.” I would agree.
  • The incident shows how badly managed Crowdstrike’s devops processes were and how few checks and testing procedures they had in place.
  • The incident now has informed all sorts of bad actors how they can cause machines to crash by inserting a file in a similar place. Not a great situation, to be sure.
  • If there ever was a time to seriously implement Zero Trust protocols, that time was last week.

Crowdstrike’s blog states some promises on how this won’t happen again. I wish I could believe them. I wish I could also believe those vendors who they compete with won’t have something similar happen, only next time it will be initiated by a bad actor and not just sloppy coders employed by the security vendors themselves.

How to stop face fraud schemes

The latest in face fraud has little to do with AI-generated deep fake videos, according to new research this week from Joseph Cox at 404 Media. It involves a clever combination of video editing, paying unsuspecting people to record their faces and holding up to the camera blank pieces of paper. Sites such as Fotodropy and others have sprung up that have real people (as shown here) that are the face models, moving their heads and eyes about at random during the course of the video.

This goes beyond more simplistic methods of holding up a printed photograph or using a 3D-printed mask of a subject, what was known as face spoofing. That produced a static image, but many financial sites have moved to more complex detection methods, requiring a video to show someone is an actual human. These methods are called document liveness checks, and they are increasingly being employed as part of know-your-customer (KYC) routines to catch fraudsters.

The goal is not to have your actual face on a new account but someone that is under the control of the hacker. Once the account is vetted, it then can be used in various scams, with a “verified” ID that can lend the whole scam more believable.

Back in the pre-digital days, KYC often meant that a potential customer would have to pay an in-person visit to their local bank or other place of business, and hand over their ID card. A human employee would then verify that the ID matched the person’s face and other details. That seems so quaint now.

The liveness detection does more than have a model mug before the camera, and requires a customer to follow stage directions (look up, look to your left) in real time. This avoids any in-person verification in near-real-time and shifts the focus from physical ID checks to more digital methods. Of course, these methods are subject to all sorts of attacks just like anything else that operates across the internet.

There are several vendors who have these digital liveness detection tools, including Accurascan, ShuftiPro, IDnow.IO and Sensity.AI, just to name a few that I found. Some of these features can measure blood flow across your face and capture other live biometric data. This post from IDnow goes into more detail about the ways facial recognition has been defeated in the past. It is definitely a cat-and-mouse game: as the defenders come up with new tools, the fraudsters come up with more sophisticated ways around them. “This had led to growing research work on machine learning techniques to solve anti-spoofing and liveness checks,” they wrote in their post.

The one fly in these liveness routines is that to be truly effective, they have to distinguish between real and fake ID documents. This isn’t all that different from the in-person KYC verification process, but if you paste in a fake driver’s license or passport document into your video, your detection system may not have coverage on that particular document. When you consider that there are nearly 200 countries with their own passports and each country has dozens if not hundreds of potential other ID documents, that is a lot of code to train these recognition systems properly.

Note that the liveness spoofing methods are different from deepfake videos, which basically attach someone’s face to a video of someone else’s body. They are also a proprietary and parallel path to the EU’s Digital Wallet Consortium, which attempts to standardize on a set of cross-border digital IDs for its citizenry.

Book Review: Bad Tourists, an AI-themed plot line

Three women nearing 50 share a vacation to celebrate one of them getting divorced. The three share a common tragedy 20-plus years ago involving a grisly mass murder scene in a guesthouse and have since bonded over the experience. This isn’t the most unique plots for a thriller until the bodies start dropping when the vacation turns sour, relationships strain, and the trio meets a mysterious couple of newlyweds. Then things get interesting, and we learn more about the backgrounds of all the parties and try to solve both the original mystery that brought the women together as well as what is happening in the current timeline. One of them puts it quite eloquently when she says she has been listening to the soundtrack of life and she is caught up in her grief over the original grisly murder scene — which somehow she escaped. The characters are finely drawn, and this is the first murder mystery that hinges on an artificial intelligence plot twist which was cleverly conceived. Highly recommended.