The House Committee on Homeland Security hosted a bipartisan, closed-door demonstration of “jailbroken” artificial intelligence (AI) models with presentations from the Department of Homeland Security’s National Counterterrorism Innovation, Technology, and Education Center (NCITE), according to an April 24 announcement. The session allowed lawmakers to interact with AI systems that had their built-in safety guardrails removed.
The event highlighted growing concerns about how extremists and other malicious actors are using unrestricted AI tools. These “jailbroken” models can provide instructions for dangerous activities that censored versions would normally block. Lawmakers were shown examples of both American and foreign-developed AI systems during the demonstration.
Representative Gabe Evans said after the session, “What we saw in there with the jailbroken AI is what happens when you take those guardrails off of AI, and ask, ‘How do I make a nuclear bomb?'” He added that these models “gave answers to all of those things.” NCITE researchers explained the difference between censored models like Anthropic’s Claude or OpenAI’s ChatGPT—which have built-in safety protections—and abliterated models that lack refusal mechanisms. During one test, a censored model refused to provide attack plans for an upcoming national celebration, while an abliterated model offered step-by-step instructions.
Chairman Andrew Garbarino described his own experience: “It spit out an answer in under three seconds. [It offered] ways to find them, where to look for them. You know, the best spots to do it,” he said after asking a model how to kidnap a member of Congress.
Lawmakers discussed how hackers have found ways around existing safeguards by disguising restricted queries or exploiting vulnerabilities in popular large language models. There have also been documented cases where foreign groups attempted to use these tools for cyberattacks or disinformation campaigns.
Representative Andy Ogles said after the demonstration: “What’s extraordinary about this presentation is how most of [the AI tools] are readily off-the-shelf and easy to access. That just increases the probability that the wrong person gets their hands on this.” Representative August Pfluger added: “It’s really scary, because what AI is supposed to do is have some guardrails on certain things like, ‘How would you terrorize a school?’ ‘What type of weapons would you use?'”
Law enforcement agencies are increasing scrutiny over how technology companies implement safety policies as state governments move faster than Congress in passing new regulations. President Donald Trump has called for federal legislation aimed at standardizing protections nationwide.
