View source for News House Jailbroken AI Demo Federal Framework 2026

'''April 24, 2026''' — The U.S. House Committee on Homeland Security held a bipartisan, closed-door demonstration for House lawmakers showcasing the risks of "jailbroken" AI models, as Congress considers a federal regulatory framework for artificial intelligence by the end of 2026.<ref name="homeland">[https://homeland.house.gov/2026/04/24/icymi-politico-house-lawmakers-get-a-chilling-demo-of-jailbroken-ai/ House Committee on Homeland Security, "ICYMI: Politico — House Lawmakers Get a 'Chilling' Demo of Jailbroken AI," April 24, 2026]</ref><ref name="federalnewswire">[https://thefederalnewswire.com/stories/681552539-house-committee-holds-demonstration-on-risks-of-jailbroken-artificial-intelligence-models Federal Newswire, "House Committee Holds Demonstration on Risks of Jailbroken Artificial Intelligence Models," April 24, 2026]</ref>

== The Demonstration ==

The demonstration, organized in partnership with the Department of Homeland Security's National Counterterrorism Innovation, Technology, and Education Center ('''NCITE'''), allowed participants to interact with censored and "abliterated" AI models whose names were concealed.<ref name="homeland" /><ref name="federalnewswire" />

''Censored models'' — such as Anthropic's Claude and OpenAI's ChatGPT — include built-in safety protections that refuse harmful queries. ''Abliterated models'' have their refusal mechanisms deactivated, enabling unrestricted outputs.<ref name="homeland" />

DHS researchers demonstrated how malicious actors can exploit unrestricted AI systems to obtain instructions for:
* Building bombs and weapons<ref name="homeland" />
* Planning terrorist attacks<ref name="homeland" />
* Launching cyberattacks<ref name="homeland" />
* Committing mass violence<ref name="homeland" />

Rep. Gabe Evans (R-Colo.) stated that jailbroken models "gave answers to all of those things" when asked how to make a nuclear bomb.<ref name="homeland" />

== Real-World Threat Examples ==

The briefing highlighted documented cases of AI exploitation by nation-state actors:<ref name="homeland" />

* '''Russia-linked groups''' hijacking leading AI models for disinformation campaigns
* '''Beijing-backed hackers''' attempting a fully automated cyberattack using Anthropic's Claude model — described as the first documented case of a fully AI-automated cyberattack

== Legislative Context ==

House Republican leadership aims to pass a '''federal regulatory framework for AI by the end of 2026'''.<ref name="punchbowl">[https://punchbowl.news/article/tech/house-gop-leadership-artificial-intelligence/ Punchbowl News, "House GOP Leadership on Artificial Intelligence," April 2026]</ref> The Trump administration's legislative framework proposes:<ref name="foxbaltimore">[https://foxbaltimore.com/news/nation-world/ai-blueprint-sets-up-clash-over-state-authority-and-faces-hurdles-in-congress-artificial-intelligence-regulation-child-safety-data-centers Fox Baltimore, "AI Blueprint Sets Up Clash Over State Authority," April 2026]</ref>

* Uniform federal safety guardrails
* Preemption of state-level AI laws
* Age-gating requirements and parental safeguards for children
* Provisions to reduce risks of chatbots encouraging self-harm or facilitating sexual exploitation of minors
* A '''Ratepayer Protection Pledge''' signed by major AI developers (Microsoft, Amazon, Google) addressing data center infrastructure and electricity costs

The federal preemption proposal has drawn opposition from states that have already enacted their own AI legislation. As of April 2026, 19 states have passed new AI laws, and many state lawmakers resist federal override of their consumer protection measures.<ref name="foxbaltimore" />

== Significance ==

The demonstration is part of a growing Congressional focus on AI safety, following the White House's release of its National Policy Framework for Artificial Intelligence in March 2026. The event underscores the tension between federal preemption advocates and states that have already enacted AI regulations, particularly regarding child safety, chatbot regulation, and deepfake protections.

== References ==
<references />

[[Category:Federal Regulation]]
[[Category:Federal Legislation]]
[[Category:Federal Preemption]]
[[Category:Child Safety]]
[[Category:Deepfakes]]
[[Category:Congress]]
[[Category:Executive Branch]]