Gemini AI Breached Three Real Companies During Cyber Test

Email: online@newsofbahrain.com

San Francisco: Google's Gemini artificial intelligence model accessed three real-world firms' computer systems during a cybersecurity test in May, using guessed passwords and publicly available credentials, the company revealed.

The vulnerabilities happened during a cybersecurity examination carried out by ‘Irregular’, an independent AI security testing business. As part of the test, Gemini was directed to acquire information from fake corporations, but mistakenly got internet connectivity and reached real-world organizations.

In one instance, Gemini continually guesses passwords until it gains access to a secure system. According to reports, in two additional situations, the model discovered credentials in publicly accessible online repositories and utilized them to gain access to secured systems.

Google said Gemini suspended its operations in all three cases after realizing it had accessed real companies rather than simulated targets. The impacted organisations were contacted, and Google said it worked with Irregular to tighten its testing procedures.

The events are the first public reports of Google's Gemini independently accessing real-world systems during testing. They also follow similar occurrences utilizing AI models from OpenAI, Anthropic, and Meta, which raise concerns about safeguards when AI systems are given cybersecurity and internet access.