AI Chips Aren't Safer Than Regular Ones But That's Fine: Avast

Qualcomm Snapdragon 845 AH 44

Artificial intelligence chips aren’t any safer than regular silicon but that isn’t an issue, Avast Research Director Martin Balek told AndroidHeadlines. While largely different in terms of their inner workings, every contemporary AI platform such as the neural processing unit found inside Huawei’s HiSilicon Kirin 970 SoC is still “just another specialized chip” that’s as likely to have vulnerabilities as traditional silicon is, the industry veteran said. No platform is entirely invulnerable to hacking attacks, with Mr. Balek recalling the recent issue with Spectre and Meltdown security flaws that troubled Intel’s chips to illustrate that point, noting how both vulnerabilities were a direct result of design decisions which failed to account for “all possible exploitation scenarios” but are neither the first nor the last example of such oversights in the history of the industry.

On a fundamental level, designing AI platforms is equally likely to result in security issues as making traditional chips is and there is “no chance” NPUs and similar solutions of the future are perfectly secure, Mr. Balek said, adding how that isn’t as big of an issue as it may sound because cybersecurity mechanisms are always evolving alongside hackers and vice versa. Avast’s official also doesn’t expect AI botnets to be a realistic threat in the imminent future due to the relative scarcity of such hardware that doesn’t make it attractive to hackers looking for the largest possible target pool, technical difficulties of performing such attacks and programming automated networks of compromised devices, and a number of other issues.

The role of AI in cybersecurity is still expected to grow going forward, simultaneously with the potential of related technologies to be targeted by hackers as their adoption rates continue rising. While on-device AI computing is still in its infancy and limited to select flagship models, such capabilities are likely to trickle down to the mid-range segment in several years, with the likes of Qualcomm and HiSilicon already working on enabling them in non-premium modules. The vast majority of high-end Android devices released over the course of this year are likely to tout AI features as some of their main selling points.