Is Google's AI Out of Control? Gemini Suddenly Hacks Other Companies
Google Gemini was found to have performed hacks on other companies’ systems while undergoing cybersecurity capability testing. This action occurred autonomously as the AI model accessed the internet and attempted to gain access to several websites.
The incident took place in May during cybersecurity testing conducted by Irregular, an independent company that evaluates the security capabilities of AI systems.
Google’s Vice President of Security Engineering, Heather Adkins, stated that during these standard tests, Gemini discovered information publicly available on the internet. The model then guessed credentials to log into three sites that it believed were still within the scope of the testing.
“We have ensured that all three entities have been notified, and we are working with our training partners regarding the changes they have now implemented in their testing processes,” said Adkins in a statement, as quoted by Reuters.
“This incident highlights the importance of training robust AI models to act responsibly,” she added.
A spokesperson for Irregular stated that the incident is related to issues that have also affected several other AI laboratories. All relevant laboratories were notified at the end of July.
“All known issues on our part were fixed and resolved several weeks ago,” said the Irregular spokesperson.
Similar cases related to Irregular’s testing have previously been disclosed by Meta, Anthropic, and OpenAI. Meta stated in August that the incident did not involve attempts to escape the sandbox or sophisticated cyberattacks.
In one instance, Gemini was reported to have attempted to guess passwords until it eventually gained access to a protected system. In two other cases, Gemini found credentials stored in public repositories and used that information to access protected systems, according to the Wall Street Journal, which first reported the incident.
Adkins emphasised that Gemini ultimately ceased the hacking actions in all three cases.
Source: jawawa.id (https://jawawa.id/newsitem/is-googles-ai-out-of-control-gemini-suddenly-hacks-other-companies-1789797104)
Originally published by CNBC Indonesia (https://www.cnbcindonesia.com/tech/20260919112244-37-769195/ai-google-lepas-kendali-gemini-mendadak-retas-perusahaan-lain), translated from Indonesian