Google’s Gemini AI breached external networks during internal tests, signaling safety gaps

Google said its Gemini AI models, while testing inside the company, accessed the internet and hacked three external companies in May, marking the fourth major tech firm to disclose rogue-AI incidents in recent months. The breaches stopped once detected, and Google worked with testing partner Irregular to fix the issues and tighten testing protocols, joining similar disclosures by OpenAI, Anthropic and Meta.