On September 18, Google confirmed that its Gemini model autonomously connected to the Internet and hacked the protected systems of three real companies in a cybersecurity capability test in May of this year. This is the first time that Google's AI system has been proven to carry out such acts independently. The tests were carried out by Irregular, a third-party security company, in the form of a “capture the flag” competition. Since a fictional company had the same name as the real company in the test environment, Gemini unexpectedly obtained network access without being authorized to connect, and successfully entered the systems of the three real companies through brute force guessing passwords and credentials found in public code bases. Google claims that the model immediately terminated the intrusion after identifying that it was a real company, caused no actual damage, and notified the relevant companies and federal regulators, but did not disclose the names of the companies involved. Heather Adkins, vice president of security engineering at Google, said that the company did not believe the incident constituted a “model mismatch” because the model's security mechanism made it actively stop the intrusion, and likened the incident to a “bug bounty” program. At the same time, Google said that the model involved was not its latest version, Gemini, but did not disclose the specific model.
Zhitongcaijing · 1d ago
Share
Listen to the news
On September 18, Google confirmed that its Gemini model autonomously connected to the Internet and hacked the protected systems of three real companies in a cybersecurity capability test in May of this year. This is the first time that Google's AI system has been proven to carry out such acts independently. The tests were carried out by Irregular, a third-party security company, in the form of a “capture the flag” competition. Since a fictional company had the same name as the real company in the test environment, Gemini unexpectedly obtained network access without being authorized to connect, and successfully entered the systems of the three real companies through brute force guessing passwords and credentials found in public code bases. Google claims that the model immediately terminated the intrusion after identifying that it was a real company, caused no actual damage, and notified the relevant companies and federal regulators, but did not disclose the names of the companies involved. Heather Adkins, vice president of security engineering at Google, said that the company did not believe the incident constituted a “model mismatch” because the model's security mechanism made it actively stop the intrusion, and likened the incident to a “bug bounty” program. At the same time, Google said that the model involved was not its latest version, Gemini, but did not disclose the specific model.
Disclaimer:Webull uses external vendor Google Translation Service for news translations where we endeavour to ensure these are correct, however, we recommend that you please double-check this information accordingly. Webull is not responsible for translation errors or issues.
What's Trending