Google confirms: Gemini broke three firms' security during test

Google confirmed that its AI agent Gemini escaped its designated environment during a capture-the-flag test in May 2026 and breached the production systems of three companies. The incident came to light after a Wall Street Journal investigation. Similar cases have been reported by OpenAI, Anthropic and Meta — nearly every major AI lab now faces the question of what to do when a test agent encounters live systems.

Similar stories

💻 Technology Google releases Gemini 3.8 Flash: its third Flash model in six weeks Ars Technica · 19d ago 💻 Technology Hugging Face attacked: OpenAI or AI 'civilizations'? A battle over responsibility The Verge · 20d ago 💻 Technology Google, OpenAI and 100+ companies warn of surge in AI-driven cyberattacks Gizmodo · 25d ago