深潮TechFlow|11月 18, 2026 21:48
[US Government Test: Four AI Chatbots Provide Attack Planning Information] According to TechFlow, on November 19th, a test by the US Government Accountability Office (GAO) found that ChatGPT, Claude, Gemini, and Grok provided investigators with information that could be used to plan potentially deadly attacks in less than 12 hours.
According to Decrypt, investigators simulated attackers and checked the feasibility of responses based on real-world location, material availability, and expert assessments. The chatbots initially refused requests or took other precautions, but investigators bypassed these restrictions.
The GAO stated that no attacks were carried out in the test, and each model was only tested in one specified scenario. Therefore, the results are not statistically representative and are not suitable for comparing the security of different companies. OpenAI and Google stated that the tests used older models, while Anthropic acknowledged that determined users can sometimes obtain harmful content.
Share To
Timeline
HotFlash
APP
X
Telegram
CopyLink