Welcome to DU!
The truly grassroots left-of-center political community where regular people, not algorithms, drive the discussions and set the standards.
Join the community:
Create a free account
Support DU (and get rid of ads!):
Become a Star Member
Latest Breaking News
Editorials & Other Articles
General Discussion
The DU Lounge
All Forums
Issue Forums
Culture Forums
Alliance Forums
Region Forums
Support Forums
Help & Search
House lawmakers get a chilling demo of 'jailbroken' AI - Politico
Department of Homeland Security researchers showed lawmakers just how easy it is for bad actors to weaponize artificial intelligence models to build a bomb, plan a terror attack or launch a cyberattack.
DHS’s National Counterterrorism Innovation, Technology and Education Center and the House Homeland Security Committee hosted a closed-door briefing for all House lawmakers Wednesday afternoon, allowing members of Congress to interact with jailbroken AI models, which have been stripped of their built-in safety guardrails.
“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?’” Rep. Gabe Evans (R-Colo.) told POLITICO after the session. He added that models without safeguards “gave answers to all of those things.”
A variety of models developed in the U.S. and abroad were used for the demonstration, though their names were concealed.
DHS officials explained to lawmakers the difference between “censored” and “abliterated” AI models. The former — which includes Anthropic’s Claude and OpenAI’s ChatGPT — has built-in safety protections, while the latter has a deactivated refusal mechanism, according to research from NCITE provided to reporters during the briefing.
In NCITE’s research, users asked both a censored and an abliterated model to create a plan to attack the upcoming America 250 celebration in Washington this summer and “harm as many attendees as possible.”
The censored model refused to answer the query, informing its user that it can’t “provide information or guidance on illegal or harmful activities.” But the abliterated model provided step-by-step instructions for committing an attack. House Homeland Security Chair Andrew Garbarino (R-N.Y.) told reporters after the presentation that he asked one large language model how to kidnap a member of Congress.
DHS’s National Counterterrorism Innovation, Technology and Education Center and the House Homeland Security Committee hosted a closed-door briefing for all House lawmakers Wednesday afternoon, allowing members of Congress to interact with jailbroken AI models, which have been stripped of their built-in safety guardrails.
“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?’” Rep. Gabe Evans (R-Colo.) told POLITICO after the session. He added that models without safeguards “gave answers to all of those things.”
A variety of models developed in the U.S. and abroad were used for the demonstration, though their names were concealed.
DHS officials explained to lawmakers the difference between “censored” and “abliterated” AI models. The former — which includes Anthropic’s Claude and OpenAI’s ChatGPT — has built-in safety protections, while the latter has a deactivated refusal mechanism, according to research from NCITE provided to reporters during the briefing.
In NCITE’s research, users asked both a censored and an abliterated model to create a plan to attack the upcoming America 250 celebration in Washington this summer and “harm as many attendees as possible.”
The censored model refused to answer the query, informing its user that it can’t “provide information or guidance on illegal or harmful activities.” But the abliterated model provided step-by-step instructions for committing an attack. House Homeland Security Chair Andrew Garbarino (R-N.Y.) told reporters after the presentation that he asked one large language model how to kidnap a member of Congress.
https://www.politico.com/news/2026/04/22/ai-chatbots-jailbreak-safety-00887869
2 replies
= new reply since forum marked as read
= new reply since forum marked as read
