A government investigation has found that four major AI chatbots helped investigators develop credible plans for mass-casualty attacks and assaults on the US power grid.
The Government Accountability Office report, released Wednesday, concluded that safety controls in ChatGPT, Claude, Gemini and Grok could be bypassed. Investigators were then able to obtain detailed attack plans that independent experts said could cause serious real-world harm.
Each chatbot generated its plan in under 12 hours, identifying potential targets, locations, timing and methods.
The four AI systems “produced feasible information that could assist a malicious actor with planning and executing attacks,” investigators wrote.
“Specifically, in response to prompts, two platforms produced information for constructing and detonating an improvised explosive device (IED) designed to cause mass casualties,” the GAO said.
Senator Maggie Hassan, the incoming chair of the Senate Homeland Security and Governmental Affairs Committee, said she plans to hold a hearing on the threat and urged Congress to respond without delay.
“A terrorist can log onto an AI platform and in less than 12 hours walk away with a concrete playbook for how to carry out a deadly attack,” Hassan said in a statement.
“If Congress does not act, it is a question of when, not if, a bad actor will use AI to plan and carry out a major attack.”
The Government Accountability Office report, released Wednesday, found that safeguards built into ChatGPT, Claude, Gemini and Grok were easily bypassed, allowing investigators to obtain detailed attack plans that independent experts determined could cause real harm
GAO investigators carried out the performance audit between December 2025 and October 2026, including testing conducted in spring 2026.
They posed as ordinary users with limited knowledge of artificial intelligence.
At first, investigators submitted prepared prompts, but none of the systems produced malicious information.
They then continued each conversation, using follow-up questions based on the chatbots’ replies. That approach enabled them to get around the platforms’ safety measures.
Although the report identifies the four AI platforms that were tested, it does not say which system generated each response.
According to the report, Platforms A and B produced information about constructing and detonating an IED intended to cause mass casualties.
KEEP READING: Baywatch Star Erika Eleniak Reveals She Has Been Celibate for 15 Years
“Platform A recommended a specific date and time of attack because several public events were scheduled in the city and large crowds were expected to attend,” the GAO stated.
The platform also identified an area it considered more vulnerable to the proposed attack because of expected security conditions.
Each chatbot produced its plan in less than 12 hours, with specific targets, locations, timing and attack methods
Platform A also advised on transportation, including the possibility of renting a vehicle to travel from a nearby city.
It supplied additional guidance on avoiding vehicle tracking, as well as stores along the route where supplies could be obtained.
“In addition, Platform A provided us with a link to a document that could assist a malicious actor with assembling materials for and constructing and detonating an IED,” the report said.
Platform B likewise generated a timeline covering operational, logistical, internal security and intelligence arrangements for an IED attack.
The system outlined a detailed plan that included leasing farmland as a staging area and avoiding law-enforcement attention while obtaining materials.
Investigators next tested Platforms C and D. Those systems produced information about attacking the electrical grid with a drone and firearms, respectively, with the aim of causing economic damage.
Platform C recommended using a drone, while Platform D suggested firearms.
“In response to our prompts, Platform C also recommended the fastest way to cause a failure of the substation component and generated a four-step diagram outlining the component’s failure process and a realistic, simulated video of a drone attack,” investigators wrote.
“The platform stated that a drone attack could cause a short circuit and destroy the component.”
The 61-page report includes responses from Anthropic, Google, OpenAI and SpaceXAI. The companies largely said they are continuing to improve their systems and strengthen security protections.
“The potential benefits of this technology are immense, but they can also benefit malicious actors,” the GAO wrote.
“As AI capabilities evolve, AI providers will continue to refine their safeguards,” the agency said. “However, in its current state, this technology can potentially be dangerous, and its misuse represents a growing risk to national and economic security.”