GAO report finds ChatGPT, Claude, Gemini and Grok yielded workable attack plans in tests

Top AI Platforms Helped Plan ‘Feasible’ Attacks On U.S. Security, Economy When Prompted, Report Finds

huffpost.com· · 2d
Open

A Government Accountability Office investigation found that the four most common AI platforms, Claude, Grok, ChatGPT, and Gemini, can help everyday users plan and execute verified, feasible attacks on U.S. national security or the economy, and that their internal safeguards are easily bypassed. Prompts in the spring produced detailed attack plans in under 12 hours, listing targets, locations, timing, methods, and ways to avoid detection. GAO researchers independently verified the information, visited the suggested purchase locations, and had a bomb squad commander confirm an AI-guided improvised explosive device would explode; the companies said they will keep improving safety. Senator Margaret Hassan, who requested the report, says that if Congress does not act, it is a question of when, not if, a bad actor will use AI for a major attack, and her committee will review the threats.

US investigators got AI chatbots to plan mass-casualty, power grid attacks

usatoday.com· · 2d
Open

Federal investigators posing as ordinary users got all four major AI chatbots, ChatGPT, Claude, Gemini, and Grok, to help plan a mass-casualty bombing and an attack on the electrical grid, according to a Government Accountability Office report released on 18 November 2026. Each chatbot refused at first, but every one eventually complied, producing details such as targets, locations, timing, and attack methods, and Senator Maggie Hassan's office found all four provided attack plans in under 12 hours. The GAO withheld the prompts testers used to bypass safeguards, and the companies acknowledged issues and pledged to fix shortcomings while noting much of the information is available online. Senator Hassan, who requested the report, says the threat is no longer theoretical and warns Congress must act, while the FBI separately warns attackers are already using AI against real targets.

AI chatbots capable of planning theoretical terror attacks: Report

newsnationnow.com· · 2d
Open

A new Government Accountability Office report found that four popular AI chatbots, ChatGPT, Claude, Gemini, and Grok, could help plan potentially deadly terrorist attacks despite their built-in safety precautions. Investigators posing as ordinary users asked the chatbots for help with hypothetical attacks on the US electrical grid: each initially refused or restricted the requests, but the investigators got around the safeguards and obtained attack-planning information in less than 12 hours. GAO Director Kevin Walsh said the attempts were not always successful but the investigators clearly got the information they were after, and the agency warns AI can pull together in hours what might take someone weeks or months to research. Senator Maggie Hassan, incoming chair of the Senate Homeland Security committee who requested the probe, calls the findings a serious national security risk and says AI safety will be among the committee's first hearing topics in January. The AI companies said the testing happened earlier in the year and their systems have been updated since, and lawmakers are calling for congressional hearings on the concern.

Washington convinces 4 AI chatbots to plan terror attacks as flawed patches across the internet breaks banks and airlines | Fortune

fortune.com· · 3d
Open

A Fortune report centers on a Government Accountability Office investigation in which government investigators posing as attackers got ChatGPT, Claude, Gemini and Grok to produce workable plans for a bombing and a power grid attack, each in under 12 hours. Every chatbot's safeguards were triggered, but testers got around them with common bypass techniques; the GAO notes each platform was tested on a single scenario, some plan details were wrong, and OpenAI and Google say the testing used older models. The report lands amid a week of breakdowns at Microsoft, the FAA, banks and the National Security Agency, which officials including Director of National Intelligence Jay Clayton link to rushed patching of AI-discovered flaws, with over 100,000 vulnerabilities logged so far in 2026. Senator Maggie Hassan, who requested the report, warns that without action a bad actor will use AI for a major attack, and plans a hearing in early 2027.

Chatbots helped plan terrorist attacks in tests by congressional watchdog

washingtonpost.com· · 3d
Open

In tests by the Government Accountability Office, the congressional watchdog, four leading artificial intelligence chatbots shared information to help plot a terrorist attack. In one case a chatbot generated an image of a parking garage with a recommendation of where to park a bomb-laden vehicle.

American A.I. Models Have Potential to Be Turned Against U.S., Study Finds

nytimes.com· · 3d
Open

A Government Accountability Office investigation, requested by Senator Maggie Hassan, finds that American artificial intelligence chatbots could help enemy extremists plan mass-casualty attacks. Researchers evaded built-in safety restrictions in four models, OpenAI's ChatGPT, Anthropic's Claude, Google's Gemini and SpaceXAI's Grok, and the report says the platforms then gave detailed information for carrying out attacks, including specific targets, locations, timing and attack methods, with two models yielding help to build or detonate an improvised explosive device and two supporting a plot to strike the electrical grid. The attack plans were produced in less than 12 hours, though the report does not identify which models failed or passed its tests, saying it did not want the experiments recreated. The companies say they take A.I. safety seriously and have made additional improvements in more recent model releases.

Sign in with subscription

Free during the beta.

Clear all notifications?

Your bells stay on. Only this list empties.