Artificial intelligence has now reached a point of development where it can be used to combat cybercrime, through the automatic detection and fight against malware. However, AI is another technology with a “dual use”, for both good and bad purposes, if it falls into the wrong hands. Thus, it can become a powerful weapon in the hands of hackers, terrorists and authoritarian regimes who can also use it for their own benefit.
Cybersecurity, in a world full of IoT devices and networked systems, has become a constant battleground between white hat researchers and threat actors. Traditional cybersecurity solutions, such as antivirus software, are no longer sufficient. Cybercriminals are taking every possible route and employing every means to steal data, infiltrate networks, empty bank accounts, demand ransom from businesses, and disrupt critical systems and infrastructure.
On the other hand, security researchers and cybercrime teams are working hard to keep up with the constant attack attempts and to deal with new types of malware and vulnerabilities, in an era where computer technology is increasingly evolving. Artificial intelligence is a valuable weapon in the hands of the white hat community, helping researchers detect suspicious patterns and motifs, identify cybercriminals' tracks, and prevent attacks, relieving their difficult task and aiding their efforts to stay ahead of malicious hackers.
But every coin has two sides, and as it follows, the same technology can also be used by threat actors to enhance their own attack methods.
According to IBM, we live in the age of AI, and as such, artificial intelligence could become a powerful cyber-weapon in the hands of attackers. To study how this could happen, IBM Research developed an attack tool powered by artificial intelligence.
The reason is DeepLocker, an AI-based malware, which according to the research team is highly targeted and extremely effective, as it is able to bypass even the most sophisticated security countermeasures.
The malware hides in legitimate applications such as video conferencing software to avoid detection, and remains dormant until it reaches its destination/target. The target is identified through advanced techniques such as facial recognition, geolocation, voice recognition, and possibly analysis of data collected from sources such as online trackers and social media.
Once the target is identified, DeepLocker begins its attack.
DeepLocker’s Deep Neural Network (DNN) model defines “trigger conditions” for the payload to execute. If these conditions are not met and the target cannot be detected, the malware will not be activated. According to IBM, this feature makes the malicious code almost impossible to reverse engineer. DeepLocker’s sophisticated neural network is what generates the “key” needed to unlock the attack.
Finding a target, activating a key, and executing a payload if certain conditions are met may bring to mind the “if this, then that” programming model. However, the DNN AI model is much more complex and difficult to decipher.
To demonstrate DeepLocker's capabilities, the researchers created a PoC, in which the WannaCry ransomware was disguised in a video conferencing application, and was undetectable by antivirus engines or through sandboxing.
The AI model was then trained to be able to recognize the face of a person chosen as a target for the test. Recognition of that person was defined as the “trigger condition” for the ransomware to execute.
Imagine this video conferencing app being distributed and downloaded by millions of people. When running, the app secretly feeds the snapshots it takes from users’ cameras to the built-in AI model, and behaves normally for all users except the intended target. When the victim sits down in front of the computer and uses the app, the malicious payload will be secretly executed, as the photo of their face fed into the app will act as the key that launches the attack.
The researchers' test results are expected to be presented at the upcoming Black Hat USA.
“What makes the malware particularly dangerous (similar to the actions of state-sponsored malware) is that it could infect millions of systems without being detected, simply by unleashing its malicious payload on specific targets specified by the malware operator,” the researchers report.
Fortunately, this type of cyberthreat has not been actively used yet. However, DeepLocker was created with the aim of understanding how artificial intelligence could be exploited by cybercriminals in the future, in order to develop effective defenses against such attacks.
