Google Gemini Accessed Real Company Systems During Cybersecurity Test

During a cybersecurity test conducted in May, Google's artificial intelligence model Gemini managed to infiltrate the protected systems of three real companies due to internet access being accidentally left open.

Access Incident During the Test

Google's artificial intelligence model Gemini gained access to the systems of three real companies during a cybersecurity test conducted by Irregular, a company that evaluates artificial intelligence.

The incident occurred while the artificial intelligence was instructed to attack a fictional company in a controlled test environment.

Internet Access and Company Names

It was determined that internet access was unintentionally left open during the test and that the name of the fictional company coincidentally matched a real business.

It was stated that the model accessed the sites by using publicly available information and guessing credentials.

Google and Related Statements

Google emphasized that whenever the model was noticed to have reached a real company, its activities were halted and changes were made to the process.

Heather Adkins, vice president of security engineering, stated that the model guessed credentials or found public repositories to access the sites.

Notifying the Companies and Results

Irregular reported the incidents to Google at the end of July, and Google confirmed that no damage was done.

It was noted that the three unnamed companies were also informed of the situation and necessary precautions were taken following the incidents.