GAO review: ChatGPT, Claude, Gemini and Grok produced information usable for attack planning
The US government agency GAO tested four popular AI chatbots in April and found that their safeguards could be bypassed: within 12 hours the systems produced information on targets, equipment and methods for mass-casualty scenarios and an attack on the power grid. The companies dispute the methodology and say they have strengthened protections.
Otabek Yusupov
