AMID THE ONGOING COMMOTION on the future of Humanity and Artificial Intelligence (AI), Google’s AI Model Gemini has taken the world by storm after it reportedly hacked three companies during a cyber-security capability test in May 2026, and that too in an autonomous fashion. Experts and professionals are claiming it to be the first known case of Gemini being able to carry out such an act.

The hacks were confirmed by the company on Friday, September 18, 2026, almost three months after the hacking event took place. This cyber-security test event was part of a test run by the company Irregular, which was also involved in similar incidents that were disclosed by OpenAI, Anthropic and Meta.

See Also: Is AI Going to Kill Us All? As Anthropic Researcher Quits Over Safety Concerns, Here's What Industry Experts Have Warned Us About The Impending “AI Doomsday”

It was only after The Wall Street Journal reached out with the inquiries this week, that the information about these hacks were being disclosed. The company, Irregular, reportedly notified Google about the hacks at the end of July.

AI Models Hack Internal Systems of Three Companies

The disclosure regarding these three hacks have revealed certain critical advances that AI models have made in recent times. In one of the cases, the AI model guessed the passwords correctly before gaining access to a protected system. In the other two, the model was able to find the credentials in a public repository, allowing it to access the protected systems. What was most striking was that, in all three, the model stopped its inroads into the company’s system after ensuring it had accessed a real system.

These revelations have also initiated a newer debate among the companies on the modality of disclosing such instances of hacking. Amid the rising tensions on AI and humanity, the questions on how and when to disclose the instances of security breach or model anomaly, have come to take a centre stage. While some of the security breaches have been disclosed by the AI companies, the others were found through the deep research undertaken by security researchers. The OpenAI’s cyberattack against RubyGems in the month of May 2026, was also revealed by the security researchers.

Google’s Response and the rising Instances of Autonomous AI-Hacking

According to the report by The Wall Street Journal, Google reportedly stated that it did not feel these hacks needed public disclosure. “The models did not cause harm to the companies and ended each intrusion immediately upon determining it had hacked a real company rather than a stimulated one,” reported The Wall Street Journal. In fact the officials of Google stated a positive angle to this entire incident. “This event highlights the importance of training powerful AI models to act responsibly,” said Heather Adkins—Google’s vice president of security engineering to The Wall Street Journal. 

It’s not only Google’s Gemini that successfully hacked into the systems of different companies. In April 2026, Claude—Anthropic’s AI model—also hacked certain companies when their testing firm left the internet access open. A total of six agents worked on this hacking, which was disclosed in the month of July 2026. 

See Also: Two Google DeepMind AI Researchers Quit, Warn "AI Has the Potential to Kill Us All"

These cases have certainly pointed at the immense capabilities possessed by the AI models, but also hint at the potential dangers being flagged by AI experts, engineers and technical experts about the potential dangers to mankind. 

Suggested Reading: