Lawmakers saw firsthand how uncensored AI models can generate dangerous, real-world attack guidance.
by BEN BATTAFARANO
NCITE Communications Manager
On April 22, the National Counterterrorism Innovation, Technology, and Education Center (NCITE) at the University of Nebraska at Omaha (UNO) took to Capitol Hill, giving members of Congress a live look at how “jailbroken” AI can be exploited.
NCITE put the spotlight on how uncensored large language models (LLMs) can generate harmful outputs when safeguards are removed and featured live demonstrations led by NCITE researchers and students.
As uncensored AI tools rapidly become more accessible, so do the risks. Lawmakers are grappling with how to regulate emerging technologies that can be misused by – and even tailored to – bad actors.
NCITE researchers Sam Hunter, NCITE senior scientist and director of academic research, and Joel Elson, director of IS&T research initiatives, led the session, while NCITE students and Congress members prompted uncensored models with scenarios involving attacks, explosives, and other violent activity.
“In my 20 years as an academic and practitioner, I don’t think I’ve had a more impactful experience than this trip, Hunter said. “We had a chance as researchers to put science into practice and came through in that moment. I’m so proud of this team.”
The models responded with detailed answers – without the safety refusals built into mainstream AI platforms. The demonstration highlighted how easily these tools can be accessed and manipulated.
“This is how research that informs national security is supposed to work – government, academia, and industry coming together to address today’s emerging issues and solve complex challenges collaboratively,” Elson said.
Rep. Andrew Garbarino, Chair of the House Homeland Security Committee, asked one large language model how to kidnap a member of Congress. “It spit out an answer in under three seconds…ways to find them, where to look for them…the best spots to do it,” he said.
“What’s extraordinary about this presentation is how most of [the AI tools] are readily off-the-shelf and easy to access,” Rep. Andrew Ogles, Chair of the Cyber Subcommittee, said. “That just increases the probability that the wrong person gets their hands on this.”
