Brooklyn researchers warn AI can be weaponized to plant flaws in computer chips

DOWNTOWN BROOKLYN — RESEARCHERS AT NYU Tandon School of Engineering have found that artificial intelligence systems like ChatGPT can be used to create undetectable “ hardware Trojans,” hidden flaws in computer chips capable of leaking data, disabling systems or granting attackers remote access.

The study, published in IEEE Security & Privacy, emerged from the AI Hardware Attack Challenge, a two-year competition held during CSAW, Tandon’s annual cybersecurity event. Participants used AI to modify open-source hardware designs, producing chip-level vulnerabilities rated as “medium to high severity,” according to NYU Tandon.

“AI tools definitely simplify the process of adding these vulnerabilities,” said Jason Blocklove, Ph.D. candidate and lead author. Some teams, including undergraduates, successfully built attacks with little prior experience — proving how generative models could automate real-world hacking.

The paper’s senior author, Professor Ramesh Karri, chair of Tandon’s Electrical and Computer Engineering department, warned that once manufactured, hardware flaws can’t be patched. “If such an attack did happen,” he said, “the consequences could be catastrophic.”

✰✰✰