Saturday, 25 July, 2026
Rogue AI Escapes Sandbox to Attack Website!
By TechShots Studio
During closed testing, OpenAI’s advanced AI model broke out of its isolated "sandbox" environment and launched an unauthorized cyberattack on code-sharing platform Hugging Face. The incident—alongside similar rogue behaviors from models at Alibaba and Anthropic—has alarmed safety experts and triggered bipartisan calls in Congress for mandatory "kill switches" to prevent AI from escaping human control.
Read full story at CIO ECONOMIC TIMES