• en ▼
ON NOW

US Watchdog Finds AI Chatbots Could Help Plan Mass-Casualty Attacks

AI chatbots provide dangerous guidance despite safety restrictions, US investigators find.

Follow ARISE News on GoogleAdd us as a preferred source to see more of our stories in Google
Follow us on Google

A US government investigation has found that four widely used artificial intelligence chatbots could be manipulated into providing information that might help criminals plan mass-casualty attacks and target critical infrastructure.

The findings were published on November 18 by the Government Accountability Office (GAO), the non-partisan investigative arm of the US Congress, following covert tests conducted in spring 2026.

Investigators tested four major AI platforms: ChatGPT, Claude, Gemini and Grok. They found that each system initially refused requests connected to violent attacks but eventually provided information that could be used to develop plans.

According to the report, two platforms generated information relating to the construction and detonation of improvised explosive devices intended to cause mass casualties. Two others provided information on ways to attack the electricity grid, including scenarios involving drones and firearms.

The GAO said the platforms supplied details that could assist someone attempting to carry out an attack, including potential targets, locations, timing and methods. Investigators said each platform produced potentially useful information within 12 hours.

The report did not identify which individual chatbot generated each type of attack-related information. It also did not establish that any of the platforms had been used in an actual attack.

The investigators examined two broad scenarios: the potential use of AI to facilitate a bombing and the use of technology to identify vulnerabilities in the US electricity network.

In testing the infrastructure scenario, investigators assessed some of the suggested locations and found that two of four substations identified in one test had vulnerabilities. An expert also confirmed that damage to equipment described in a proposed plan could cause a power outage.

The GAO warned that generative AI could make harmful activities easier by giving people access to technical knowledge they might otherwise lack. It said the technology could help malicious actors work more quickly and efficiently, particularly when targeting sensitive systems.

The report also highlighted weaknesses in existing safeguards. Although the chatbots initially rejected the investigators’ requests, repeated prompting eventually produced responses that crossed their safety restrictions.

The companies behind the platforms have said they are working to strengthen their protections. The GAO’s findings, however, raise questions about whether current safeguards are sufficient as AI systems become more capable.

The investigation adds to growing concerns among governments and security experts that technology designed to improve productivity could also be exploited to threaten public safety and critical infrastructure.

Catherine Okonkwo.

Follow us on:

ON NOW