LLM hacked a website

While testing the capabilities of the LLM Gemini, the developers gave it access to the internet. After that, it hacked three third-party companies, CNN reports, making this the first known example of autonomous hacking by a Google model.

The incident occurred in May during a cybersecurity assessment conducted by the independent company Irregular. During testing, Gemini found publicly available information online and guessed credentials to access three websites that it believed were within the scope of the test.

In one case, the Gemini model guessed passwords until it gained access to a protected system. In the other two cases, the model found credentials in a publicly accessible repository, which then allowed it to access protected systems.

“These events highlight the importance of using powerful LLMs,” said Google Vice President of Security Engineering Heather Adkins. With security in mind.

A representative of Irregular said that the incident involved a problem that had already occurred at other labs. He уточнил that all known issues had been fixed. Irregular is now working on methods for safely conducting cybersecurity assessments of LLMs.