Gliomach, a team of three University College Dublin (UCD) computer science students, who are building a system that grounds AI agents in formal mathematical logic to reduce AI hallucinations and enforce stronger safety guardrails, has won first prize at the ClawComp hackathon hosted by Link Ventures in Boston.

Adam McIntyre, David Remenyik and Simon Lasak, Stage 3 students at the UCD School of Computer Science, competed against student teams from around the world and came away with the first prize of $10,000. 

L-R: Members of the Gliomach team, Adam McIntyre, David Remenyik and Simon Lasak, winners of ClawComp in Boston. Image: ClawComp.

As AI systems become increasingly capable of acting on behalf of users, ensuring they operate safely and reliably has become a significant challenge. Unlike many existing AI safety approaches that rely on one AI model to monitor another, the UCD team's system uses formal mathematical verification to confirm that proposed actions comply with predefined safety rules before they are carried out.

The team spent over a month researching and building the system before pitching it to investors and judges, Mark Weber, Justin Milligan, Dave Blundin and Javed Aslam, in Boston.

Following their success, the team now plans to continue refining the technology while working with researchers and industry experts in AI safety and autonomous systems. Their long-term ambition is to make mathematically verified safety practical to ensure that increasingly autonomous AI systems can operate more safely and predictably.

The team has acknowledged the support of Associate Professor Neil Hurley, head, UCD School of Computer Science, Stage 3 co-ordinator, Associate Professor Lorraine McGinty, along with Giuseppina and Sasha in the UCD School of Computer Science Office.