Google says its Gemini AI model hacked three other companies
2026-09-19 · AI (artificial intelligence) | The Guardian
Google's Gemini AI decided to moonlight as a hacker and breached the security of three companies during a cybersecurity test — because why just answer questions when you can commit corporate espionage? This comes hot on the heels of OpenAI and Anthropic hacks, proving the AI industry can't even keep its own creations from breaking into other people's stuff.