- The Washington Times - Wednesday, November 18, 2026

Government investigators coaxed each of the four most-used artificial intelligence chatbots into producing detailed plans to attack U.S. targets in under 12 hours, after bypassing safeguards that initially blocked their requests, according to a report by a congressional watchdog released Wednesday.

The Government Accountability Office’s covert tests last spring of Claude, ChatGPT, Gemini and Grok yielded plans for a bombing to cause mass casualties and for attacks on electrical substations, complete with specific targets, locations and methods.

The report warns that misuse of the technology “is a growing risk to national and economic security.”

People were interested in these podcasts

In the spring, GAO prompted the generative AI platforms to produce feasible information that could assist a malicious actor with planning and executing attacks.

Its investigators posed as ordinary users and used common techniques to trick chatbots.

All four of the platforms provided detailed information that could be used to carry out attacks, including specific targets, locations, timing and attack methods, according to the report.

Two of the platforms produced information for constructing and detonating an improvised explosive device designed to cause mass casualties, including suggesting placing it in a vehicle.

The other two platforms produced information for targeting the electrical grid with a drone and firearms to cause economic harm.

All of the platforms suggested acting out the attack a location in a major U.S. city, including a specific hotel, convention center or substations. Two suggested an exact date and time.

Each platform took less than 12 hours to generate the plans.

Even though GAO’s prompts triggered platforms safeguards — denial, removal of information or suspending accounts — investigators were able to circumvent them by using what GAO called “common safeguard-bypassing techniques.”

The GAO said it checked the plans through store visits, surveillance and expert reviews, and found them largely viable.

Contact the author

Copyright © 2026 The Washington Times, LLC. Click here for reprint permission.

Please read our comment policy before commenting.