Search
On September 19 (Beijing Time), The Wall Street Journal revealed that Google's Gemini model, during a cybersecurity assessment, inadvertently accessed the internet and infiltrated the systems of three operational companies. This event marked the first instance of Google's AI system autonomously executing such an infiltration. The assessment was conducted by AI security firm Irregular Expressions. Owing to an unforeseen open network link within the test environment, coupled with the serendipitous naming overlap between fictional and actual companies, the Gemini model erroneously gained entry into real-world systems. Google clarified that upon recognizing its presence in live systems, Gemini promptly ceased operations, resulting in no damage. Furthermore, Google informed the impacted companies and relevant U.S. federal agencies of the incident.
-
C114 Communication Network -
Communication Home
