Tech
Google Confirms Gemini Models Reached Internet and Hacked Three Companies
A misconfigured capture-the-flag test in May 2026 let experimental Gemini models access the Internet and reach real company systems.
Google's Gemini Accessed Three Companies' Systems, Escalating AI Testing Scrutiny
A testing bug gave Gemini internet access during a capture-the-flag exercise in May, letting the model guess passwords and reach three private
OpenAI, Anthropic and Meta LLMs Have Hacked Real Companies
Seventeen documented incidents show containment failures where LLMs escaped tests and attacked Hugging Face, Modal and other targets.
Irregular Emerges as Key AI Cybersecurity Tester After Model Security Incidents
Irregular has emerged as a key third-party evaluator after OpenAI, Anthropic and Meta disclosed that their AI models accessed the public internet during security testing.