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.

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.

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.

DHS researchers demonstrated how malicious actors can exploit unrestricted AI systems to obtain instructions for:

• Building bombs and weapons
• Planning terrorist attacks
• Launching cyberattacks
• Committing mass violence

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

Real-World Threat Examples

The briefing highlighted documented cases of AI exploitation by nation-state actors:

• 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. The Trump administration's legislative framework proposes:

• 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.

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.