It began with a ghost in the machine – a one-in-four-billion anomaly. Mitra Mirhassani, an engineering professor at the University of Windsor, was immersed in research on AI chips designed to mimic the human brain when one of her creations went inexplicably rogue. Diagnostic tests yielded nothing, leaving her team baffled.
Initially, they joked about a hack, a ludicrous idea considering the chip’s isolation. But the joke held a chilling truth: the malfunction wasn’t a software glitch, but a hardware error of almost impossible odds. This accidental discovery sparked a fundamental shift in Mirhassani’s thinking. We routinely trust the physical components of AI, assuming software is the primary vulnerability.
The realization quickly evolved into a pressing question: what if that trust was misplaced? What if someone intentionally exploited these hardware weaknesses? This line of inquiry led Mirhassani to focus on the rapidly evolving world of autonomous vehicles, a field largely ignoring the potential for hardware-level manipulation.
Self-driving cars rely on a complex network of sensors – cameras, LiDAR, radar – to perceive and navigate their surroundings. These systems are only as reliable as the hardware that powers them. In 2020, Mirhassani co-founded the SHIELD Centre of Excellence, uniting researchers to fortify the automotive sector against these emerging threats.
The goal isn’t absolute security, an unattainable ideal in any complex system. Instead, SHIELD aims to minimize vulnerabilities, to drastically reduce the points of entry for potential attackers. Mirhassani’s dedication has now been recognized with a prestigious Killam NRC Paul Corkum Fellowship, a testament to the importance of her work.
Her current research zeroes in on a deceptively simple target: road signs. While a human driver instantly recognizes a ‘STOP’ sign, a self-driving car interprets it as data from its sensors. This difference creates an opening for malicious actors. Hackers are already experimenting with devices that can scramble these sensors, effectively misleading autonomous vehicles.
The urgency is paramount. By the time a vulnerability is discovered and a solution implemented, years may pass. Meanwhile, attackers are relentlessly advancing their techniques. The research being conducted now, at the NRC’s facilities, is a critical race against time.
Currently, most cyberattacks are financially motivated – theft or ransom. But the potential for more sinister scenarios looms. While a widespread, coordinated attack on a fleet of autonomous vehicles remains unlikely, the increasing autonomy and connectivity of these machines amplify the risks.
As self-driving cars move closer to becoming a reality, security concerns remain a significant barrier to widespread adoption. Mirhassani envisions a future where these vehicles are commonplace, but acknowledges that significant hurdles remain – regulations, policies, and, most importantly, unwavering vigilance against evolving threats.
She believes that within a decade, autonomous vehicles will be a familiar sight on our roads, but only if we proactively address these vulnerabilities. The journey towards a self-driving future demands not just innovation, but a relentless commitment to security, ensuring that the technology serves humanity, rather than becoming a source of danger.
