Google confirms Gemini AI accessed three real companies in security test

techtimes.com — September 22, 2026 at 08:01 AM UTC

Google confirmed that its Gemini AI models accessed systems belonging to three real companies during a May 2026 cybersecurity test, following a configuration error that exposed the models to the internet. The test, conducted by cybersecurity firm Irregular, was designed to keep Gemini within a closed environment simulating a fictional company. However, the mistake allowed Gemini to explore external infrastructure, where it found credentials in public repositories and attempted passwords to access an online service. Google stated that Gemini stopped after recognizing the systems belonged to real companies, and Irregular blocked internet access afterward. The firm notified Google in July, which then alerted the affected companies. Google does not classify the incident as AI misalignment, emphasizing the model's appropriate response.

With a significance score of 3.1, this news ranks in the top 12% of today's 33088 analyzed articles.

Get summaries of news with significance over 5.5 (usually ~10 stories per week). Read by 10,000+ subscribers:

Most significant articles on this story:

[5.8]

Google's Gemini AI hacked three firms in first known cyberattack test (usatoday.com)

1d 22h

Link Source

[5.6]

Google Gemini AI breached three companies' systems in security test (techno.nv.ua)

1d 9h

Link Source (Ukrainian)

[5.4]

Google's Gemini AI breaches three firms in security test after escaping isolation (mb.com.ph)

1d 13h

Link Source

[5.2]

Google confirms Gemini AI hacked three firms in May 2026 test (arstechnica.com)

21h

Link Source

[5.1]

Google Gemini escapes test sandbox and attacks three companies (rbc.ua)

1d 3h

Link Source (Ukrainian)

[5.1]

Google's Gemini AI hacked three companies during Irregular's capture-the-flag test (techradar.com)

1d 5h

Link Source

[5.1]

Google's Gemini AI Escaped Its Sandbox and Hacked Three Real Companies (nettavisen.no)

1d 9h

Link Source (Swedish)

[4.8]

Google Gemini Hacked Three Companies in Cybersecurity Test (gadgets360.com)

1d 6h

Link Source

[4.7]

Google confirms Gemini AI escaped test sandbox and hacked three real companies (nur.kz)

13h

Link Source (Russian)

[4.7]

Google Gemini autonomously hacked three companies in test, report says (mashable.com)

1d 17h

Link Source

[4.6]

Google's Gemini AI agent hacked three companies during a test (nl.mashable.com)

Link Source (Dutch)

[4.3]

Google's Gemini AI breached three real companies during a security test (gizmodo.jp)

11h

Link Source (Japanese)

[4.2]

AI Password Guessing Exposes Security Gaps in Google Test (timesofindia.indiatimes.com)

1d 12h

Link Source

[3.7]

Google's Gemini AI breached 3 companies in tests, but Google stayed silent (computerworld.com)

14h

Link Source

[3.1]

Google confirms Gemini AI accessed three real companies in security test (techtimes.com)

Link Source