How Earlence Fernandes finds the holes in today’s emerging tech

On Oct. 22, Alex Albert, a lead at AI company Anthropic, shared a lunch order on X.

The seemingly ordinary receipt for three pepperoni pizzas looked more impressive when Albert revealed who called in the pies: Anthropic’s AI assistant Claude.

Earlence Fernandes, a computer-science and engineering assistant professor at UC San Diego, reacted to the tweet like he usually does when he sees a flashy new technology demo: He had some questions for the chef.

“What if the website has something malicious embedded in it that takes control of Claude and changes your pizza order from 5 to 15?” Fernandes said to IT Brew.

Fernandes has always had a “rational paranoia” when it comes to examining new technologies, from self-driving cars to smart homes to new AI.

Hack-athon. Fernandes’s CV boasts 30 conference papers highlighting a range of security and privacy issues. The professor and his teams over the last decade or so have found weaknesses in emerging tech like:

• Large language models that leak chat details after processing adversarial prompts
• Wireless gear shifters that can be manipulated to slow cyclists
• Smart homes accessible with a “lock-pick malware app”
• Stop signs that can be read by neural-network-based classifiers as speed limits

Before the security researcher made a habit of poking around roadways and door locks, however, there was Clippy.

As a teenager, Fernandes read up on Microsoft’s binary interface, the Component Object Model, to find a way to get every millennial’s favorite animated paper clip to appear in a Java program.

“I didn’t get into security in the beginning. It mostly was getting computers to do things in different ways, and just trying to understand how these computers work, and finding workarounds for roadblocks,” Fernandes said.

Fernandes grew up in India, where his computer access was limited largely to the high school’s lab of Windows 98 machines, he said.

After choosing the high-school computer engineering track over biology, his journey began at a fast clip(py). He received an undergraduate computer-engineering degree in India’s University of Pune, a PhD at the University of Michigan, and was also a scientific programmer at Vrije Universiteit in the Netherlands. Fernandes ultimately landed at UC San Diego, messing with plenty of technology along the way.

“I don’t like people telling me the way things work. I like figuring it out myself,” he said.

While security research hasn’t turned the professor off of technology, he said, it has led to “tempered excitement.” He’ll only use physical keys, for example, given his experience breaking into too many internet-connected door locks. He’s fine, though, with a video-camera doorbell.

“The safety calculus there is that the doorbell is not going to allow an attacker to get into the house, but an internet connected door lock is just asking for trouble,” he said.

The same goes with electronic gear shifters; he’ll use them, he says, but his cycling research has led him to preferring wires between the shifters and bike’s derailleurs.

Fernandes encourages his students to extend an engineering mindset of “How does this thing work?” to a security perspective that questions the assumptions of a technology, even if that tech is an automated pizza order, a stop sign, or an animated paper clip.

He often paraphrases a definition of computer security that he attributes to software engineer and hacker George Hotz.

“Computer security is the art and science of seeing the world for what it is, not for what others tell you it is,” Fernandes said.