AI Threatens to Make Side-Channel Attacks More Effective and Achievable

Cybercriminals tend to rely on the same basic methodologies to steal data – phishing, credential theft, vulnerability exploits, etc. So it can be easy to overlook more complex, lesson common malicious practices, such as side-channel attacks.

But the advent of easy-to-use artificial intelligence tools may start to make the prospects of successful side-channel attacks more realistic. Much like AI has already helped remove the barriers for creating convincing deepfake videos and phishing messages, and much like it may help accelerate efforts to defeat decryption through quantum computing, AI may also be able to make side-channel attacks more accessible.

Side-channel attacks steal secrets by capturing and interpreting subtle measurements of targeted machines’ physical properties or outputs, such as how they emit electromagnetic radiation, generate sound or consume power. Historically, these attacks have not been pragmatic for many bad actors – though they were not necessarily out of reach for a committed cyber enemy with powerful resources.

But by leveraging the computational powers of AI to more quickly identify and decipher patterns from how a particular machine operates, is it possible that cybercriminals may soon find a world of side-channel attacks opening up to them? The possibility is worth discussing in order to increase awareness around this threat.

One institute known for uncovering and studying side-channel attacks is Ben-Gurion University of the Negev in Israel. Just this past September, the academic institution published a paper detailing a newly discovered side-channel attack exploit, dubbed Pixhell, that enables would-be attackers to leak and capture secrets from air-gapped and audio-gapped systems via the LCD screen noise of a connected computer monitor.

If one were able to infect the computer with a particular malware program (perhaps via an insider threat), the “malware in the air-gap and audio-gap computers generates crafted pixel patterns that produce noise in the frequency range of 0-22 kHz. The malicious code exploits the sound generated by coils and capacitors to control the frequencies emanating from the screen. Acoustic signals can encode and transmit sensitive information” – meaning the infected machine would be acoustically leaking secrets that someone could then capture from another nearby device.

Side-channel attacks may sound impractical, and some can be tricky to execute. But in 2017, a series of independent and academic researchers secretly collaborated on a massive effort to develop a fix for two serious side-channel vulnerabilities – dubbed Spectre and Meltdown – that were baked into billions of computer chips, thus making these devices susceptible to data theft. Only months later in January 2018 were these vulnerabilities revealed to the public. This immense undertaking demonstrates how seriously cyber experts have considered side-channel vulnerabilities in the past. And with AI having greatly evolved since then, such flaws may be more exploitable than ever before.