UM-Flint partners with NetSTAR to enhance internet safety through advanced AI techniques

The University of Michigan-Flint has announced a groundbreaking partnership with NetSTAR, a global leader in web filtering and internet security solutions. The collaboration aims to develop advanced artificial intelligence technologies to enhance internet safety, particularly for vulnerable populations like children and the elderly. The project will also spur productivity by blocking access to harmful content, protecting sensitive information and minimizing distractions in educational and work settings. 

The partnership was officially announced at a signing ceremony during the Symposium on Cybersecurity Horizons: Unveiling Pathways, Tackling Challenges, and Shaping Research & Innovation, the first in a series of annual conferences hosted by UM-Flint's Center for Cybersecurity.

Nearly 200 attendees gathered at the university's Riverfront Conference Center to see presenters from Microsoft, Amazon, General Motors and many other institutions, with each sharing insights into career pathways and industry challenges.

Khalid Malik, professor of computer science and director of the Center for Cybersecurity at UM-Flint, coordinated the event and hosted NetSTAR's leadership, including Yasuaki Arakawa, the company's chief technology officer, who worked with Malik in the past.

"I first met Dr. Malik around 2010 when we both worked at NetSTAR in Tokyo," Arakawa said. "We collaborated on malware detection and website analysis, which was very successful. Our experience together was a huge factor in NetSTAR's decision to work with UM-Flint."

After Malik moved to the U.S., the two kept in touch via LinkedIn, keeping the door open for future partnerships in the field of cybersecurity. Finally, in May 2024, Malik and Arakawa began a renewed collaboration that would focus on making the internet safer for users, especially minors. 

Their joint work focuses on AI technologies capable of real-time detection and categorization of malicious websites. The traditional method of weeding out dangerous URLs involves manually updating databases. But with millions of web addresses being created and altered by the day, the future of cybersecurity depends on AI technologies that can automatically detect harmful content like pornography or phishing sites.

According to Malik, significant hurdles to development include encrypted DNS requests, real-time categorization, evolving website content and the multimodal nature of the internet.

"In the future, many people will use the internet with encrypted DNS requests, which hide activity from ISPs," he said. "Encrypted traffic makes it difficult to categorize and filter harmful websites. We need to categorize URLs in real time and can't always trust their previously assigned categories." 

Neuro-symbolic AI could provide a solution by allowing for more sophisticated analysis and decision-making. The process stands in contrast to traditional deep learning, which, in Malik's experience, is excellent at pattern recognition but falls short in reasoning.

"Our ultimate goal is to strike a balance between internet safety and privacy," Malik said. "To address these challenges, we're developing neuro-symbolic-based techniques for URL classification, which involves a combination of symbolic reasoning and multi-modal deep AI that can process audio, video and text."

An AI model capable of accurate online threat assessment would represent a technological leap forward in the field of cybersecurity, and in his remarks at the Symposium on Cybersecurity Horizons, Arakawa underscored the importance of the project's success.

"Our goal is to use technology to create solutions that improve internet safety and security for all," Arakawa said. "But this collaboration is more than just a tech project—it's about making a real difference in the world. By combining NetSTAR's strengths with UM-Flint's exceptional skills, we aim to tackle the complex cybersecurity challenges we face today." 

To learn more, visit UM-Flint's Center for Cybersecurity webpage.