In a business email compromise, generally, the attacker uses emails and social engineering techniques to have one person with financial power in a company transfer money to a bank account the attacker owns. This kind of fraud is a sophisticated scam targeting companies and individuals who perform legitimate funds transfers.
Statistics from the FBI’s Internet Crime Complaint Center, law enforcement and filings with financial institutions indicate that BEC alone caused an exposed loss of more than $43 billion USD between 2016 and 2021.
BEC detection and blocking based on email characteristics
BEC attackers use different social engineering techniques, yet most of the time, they use emails set up to pretend to come from a legitimate person in contact with the target. To achieve that, they often register email addresses close to the legitimate one from the impersonated person.
SEE: Password breach: Why pop culture and passwords don’t mix (free PDF) (TechRepublic)
Therefore, one way to take advantage of this situation for blocking purposes might be to only allow external emails from trusted senders. Another variant consists of blocking emails coming from free email providers, as fraudsters often use those, as exposed by Cisco Talos (Figure A).
Yet, building email block lists can be difficult for users, as they sometimes get external emails from people that have not yet been added to their trust list.
Several security software options might help to deploy policy-based detections of BEC emails. Those solutions generally store the names and email addresses of executives in a database that is used on every incoming email. If the name is found in the “from” field of an email and does not match the legitimate one stored in the database, a BEC attempt alert is raised.
There is yet an obvious limitation to this detection type: If the email comes from any other person than the executive, no alert is raised. Attackers might also spoof the legitimate address in the “from” field in some cases, but use a different “reply to” field, which might help bypass some detections if they are only based on the “from” field.
And in some cases, the fraudsters might have compromised the executive email box and would be able to send emails impersonating them without raising alerts for that kind of detection.
Another approach: ML-based model profile building
According to Talos research, it is possible to build a profile of C-level executives by using a machine learning algorithm to analyze all emails.
This profile would be based on several items, such as the person’s writing style, activities, geolocation when sending emails, timestamp of posting. A relations graph capturing the person’s email interactions with others might also be generated.
In case of any deviation from the profile, a BEC alert could be raised.
Just as for traditional detection, the method has some limitations. Generating the profile needs to be done from real traffic, and data collection, model building and training will take time. Also, building it for every employee of the company would be challenging.
As for the non-executives impersonated persons in companies, Talos indicates that they are engineers more than 50% of the time (Figure B).
An intent-based approach to BEC detection
This approach aims at solving the biggest problems of the policy-based and machine learning algorithm methods: the non-scalability of the model and the difficulties of maintaining a database of sender email addresses and their names.
To overcome those limitations in detecting BEC fraud, Talos offers an intent-based approach.
This approach separates the detection of the BEC threat into two distinct problems. The first one is a binary class problem. It classifies emails into a BEC message. The second one is a multi-class problem, classifying the BEC into the type of scam.
SEE: Optimize and secure your team’s Apple devices with Jamf Now (TechRepublic Academy)
The researcher explains that the intent-based approach not only detects BEC emails but also categorizes it into a kind of BEC scam: payroll, money transfer, initial lure, gift card scams, invoice scams, acquisition scams, W2 scams, aging reports and more.
From a technical point of view, it consists of extracting the email text and converting sentences into numeric vectors. This conversion is based on NNLM or BERT algorithms, which takes the meaning of words in the sentence and then performs detection and classification using deep neural networks. The final output is a probability of the email to be a BEC attempt. A low confidence in the result will lead to more analytic detections to provide a final trust indicator.
This approach works no matter who is impersonated in the company.
The need for raising awareness
No matter what kind of automated solution is deployed to protect companies and employees from falling to BEC fraud, it is still a great addition to train employees and raise awareness on what BEC fraud is, how it happens, what kind of social engineering tricks it uses, and what should raise suspicion.
Users need also be aware that BEC fraud can happen not only by email but also by voice. Some BEC fraud might leverage phone calls to approach the employees or even SMS.
Any attempt to change a modus operandi for a financial transfer, any sudden change of a recipient banking account should immediately raise an alarm and be investigated. The user targeted should never be afraid to reach out to the sender of the request via another communication channel to confirm there is no ongoing scam.
Secure your team’s mobile devices and detect phishing scams faster with the Mobile Device Security Policy from the policy experts at TechRepublic Premium.
Disclosure: I work for Trend Micro, but the views expressed in this article are mine.