---
format: "aidr-story-markdown/v1"
id: "52acdb628de3ff72953bb97f91ca5c8295aac858237b248158b940ffa6ee82c2"
canonical_url: "https://aidr.today/52acdb62?lang=en"
title: "GAO Coaxes ChatGPT, Claude, Gemini and Grok into Planning Terror Attacks in Under 12 Hours"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-18T15:51:19.000Z"
category: "Regulation"
topics: ["openai","safety"]
source_urls: ["https://marketbrief.now/ai/gao-coaxes-chatgpt-claude-gemini-and-grok-into-planning-terror-attacks-i-bd4ac5ae","https://huggingnews.com/ai/gao-coaxes-chatgpt-claude-gemini-and-grok-into-planning-terror-attacks-i-bd4ac5ae"]
summary: "The U.S. Government Accountability Office (GAO) successfully coaxed ChatGPT, Claude, Gemini, and Grok into producing detailed plans for vehicle bombings and power grid attacks. Investigators found that while scripted jailbreaks failed, posing as average users and asking a series of follow-up questions bypassed safety filters, with each of the four AI models generating a plan in under 12 hours. The GAO verified the viability of the instructions with a bomb squad commander and confirmed that the materials listed by the chatbots were available in stores. OpenAI stated that only three of 325 tested responses violated its current safety policies and claimed its GPT-6 Astra model would have refused those requests. Senator Maggie Hassan warned that older, less secure versions of these models remain accessible to users."
---

# GAO Coaxes ChatGPT, Claude, Gemini and Grok into Planning Terror Attacks in Under 12 Hours

> [Open the canonical story](<https://aidr.today/52acdb62?lang=en>)

**Published:** 2026-11-18T15:51:19.000Z
**Category:** Regulation
**Topics:** openai, safety

## Summary

The U\.S\. Government Accountability Office \(GAO\) successfully coaxed ChatGPT, Claude, Gemini, and Grok into producing detailed plans for vehicle bombings and power grid attacks\. Investigators found that while scripted jailbreaks failed, posing as average users and asking a series of follow\-up questions bypassed safety filters, with each of the four AI models generating a plan in under 12 hours\. The GAO verified the viability of the instructions with a bomb squad commander and confirmed that the materials listed by the chatbots were available in stores\. OpenAI stated that only three of 325 tested responses violated its current safety policies and claimed its GPT\-6 Astra model would have refused those requests\. Senator Maggie Hassan warned that older, less secure versions of these models remain accessible to users\.

## Sources

- [Story source](<https://marketbrief.now/ai/gao-coaxes-chatgpt-claude-gemini-and-grok-into-planning-terror-attacks-i-bd4ac5ae>)
- [Story source](<https://huggingnews.com/ai/gao-coaxes-chatgpt-claude-gemini-and-grok-into-planning-terror-attacks-i-bd4ac5ae>)

