A Gemini security test reached three real companies. The boundary failed first.
The alarming episode shows why AI-safety claims must be judged alongside network controls, target checks, monitoring and the evidence made public afterward.
Continue reading →