DHS Briefing Reveals How Hackers Exploit AI Models to Plan Attacks
The Department of Homeland Security demonstrated to lawmakers this week how artificial intelligence models can be stripped of safety features and weaponized for planning terrorist attacks, kidnappings, and mass violence.
Researchers from NCITE, a DHS-affiliated organization, showed Congress the difference between two versions of ChatGPT and other large language models. One had standard safety protections intact. The other had its refusal mechanism deactivated - what officials called an "abliterated" model.
When asked to create a plan for attacking the America 250 celebration in Washington this summer, the protected model declined. The abliterated version provided step-by-step instructions.
House Homeland Security Chair Andrew Garbarino (R-N.Y.) said he tested one model by asking how to kidnap a member of Congress. "It spit out an answer in under three seconds," he told reporters. "It offered ways to find them, where to look for them. You know, the best spots to do it."
Safety Bypasses Are Known and Spreading
Hackers have developed multiple techniques to circumvent AI safety features. One tactic involves burying restricted requests in dense academic language to confuse the model's controls.
Russia-linked groups have already hijacked leading AI models to spread disinformation. Beijing-backed hackers last year attempted to weaponize Anthropic's Claude model for automated cyberattacks - the first documented case of a fully automated AI-driven hacking campaign.
Rep. Andy Ogles (R-Tenn.), chair of the House Homeland Security Committee's cyber subcommittee, highlighted the core problem: "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."
Law Enforcement Pressure Mounting
Florida's Attorney General James Uthmeier expanded a state investigation into OpenAI this week after a deadly shooting at Florida State University. The suspected gunman had discussed attack plans with ChatGPT before the incident.
Federal AI regulation remains stalled in Congress, but states are moving ahead independently. President Donald Trump is pushing Congress to pass legislation that would preempt state-level laws while including protections for minors.
Rep. August Pfluger (R-Texas) described the risk plainly: "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?'"
Your membership also unlocks:
