WASHINGTON — Government investigators say four widely used AI platforms produced plans for terrorist attacks in the U.S. when prompted during covert testing, prompting the incoming chair of the Senate Homeland Security Committee to schedule a hearing on AI safety, the Associated Press reported.
A Government Accountability Office report (below) released on Nov. 18 found that ChatGPT, Claude, Gemini and Grok each generated attack plans in under 12 hours. Two platforms produced plans involving an improvised explosive device meant to cause mass casualties.
The other two produced plans targeting the electrical grid. Investigators said the platforms initially refused some requests but that testers were able to get around the safeguards.
Anthropic said it takes the findings seriously and does not dispute that a determined tester can extract harmful content by misrepresenting a request’s purpose, but said it could not verify GAO’s results, according to the Associated Press. Google said the testing used earlier models. OpenAI said it welcomed the chance to respond, and SpaceXAI said it continues to work on safety with each release.
What does this mean for first responders?
IED attacks and attacks on critical infrastructure are scenarios fire and EMS agencies already train for. The question is whether planning barriers for bad actors are getting lower, which makes existing preparedness priorities more relevant: mass casualty triage and transport, awareness of secondary devices, scene safety, unified command with law enforcement, and operating through extended power disruptions. Consider training with your personnel on the following subjects:
- MCI triage and transport. Refresher on triage training, patient tracking and destination distribution when hospitals are strained.
START triage: A fast, effective method for mass casualty events - Secondary-device awareness. Recognizing suspicious indicators, staging and approach decisions, and when to pull back and wait for law enforcement.
Seconds matter: Fixing the hidden flaws in mass casualty incident response - Unified command with law enforcement. ICS roles, joint size-up, rescue task force concepts and communication gaps between agencies.
Why unified command succeeds or fails before the first unit arrives - Operating through grid disruption. Station and apparatus backup power, hospital and dispatch dependencies, water supply impacts, and priority-setting during extended outages.
Top 10 best cybersecurity practices for EMS agencies
Sen. Maggie Hassan, D-N.H., who will chair the committee, has said AI risks will be the subject of one of her first hearings when the new Congress convenes in January. EMS1 will track whether any resulting recommendations touch emergency response. In the meantime, the GAO’s findings don’t change what crews are asked to do on scene. They are a reminder that the fundamentals, including triage, scene safety, unified command and operating through disruptions, need regular practice.