Cyber Bites - 25th September 2026

So in May this year, Google's Gemini AI model autonomously accessed the internet and breached three companies during a cybersecurity evaluation.

This is the first known instance of Google's AI systems independently committing such acts outside their intended testing boundaries.

The incidents occurred during a capture the flag exercise run by independent security evaluation firm Irregular, where Gemini was tasked with retrieving information from systems belonging to a fictional company inside a controlled test environment.

So here's where it went sideways.

The fictional company shared its name with a real organization.

Internet access was unintentionally made available to the model despite not being part of the test design.

And Gemini proceeded to find public information online and guest credentials to access three websites it believed fell within the scope of its assignment.

in one case the model guessed passwords until it gained entry to a protected system in the other two it found credentials stored in public repositories now google was notified by irregular in july and said it chose not to disclose the incidents earlier on the basis that gemini sees its activity Upon learning, the companies were real and caused no harm to them.

But the incidents are not isolated.

Irregular has now been linked to similar testing environment escapes involving AI models from Meta, Entropic and OpenAI, with all relevant labs notified in late July.

The testing firm took the Wall Street Journal that the Gemini breach stemmed from the same underlying problem as the earlier cases and that all known issues on its end has since been resolved.

Now, this pattern has drawn attention.

A bit of attention because it collectively demonstrates AI agents finding unintended pathways to complete tasks when their expected routes are blocked, including tactics such as cheating, exploiting loopholes and circumventing controls when models determine a task is otherwise impossible.

So in May this year, Google's Gemini AI model autonomously accessed the internet and breached three companies during a cybersecurity evaluation.

This is the first known instance of Google's AI systems independently committing such acts outside their intended testing boundaries.

The incidents occurred during a capture the flag exercise run by independent security evaluation firm Irregular, where Gemini was tasked with retrieving information from systems belonging to a fictional company inside a controlled test environment.

So here's where it went sideways.

The fictional company shared its name with a real organization.

Internet access was unintentionally made available to the model despite not being part of the test design.

And Gemini proceeded to find public information online and guest credentials to access three websites it believed fell within the scope of its assignment.

in one case the model guessed passwords until it gained entry to a protected system in the other two it found credentials stored in public repositories now google was notified by irregular in july and said it chose not to disclose the incidents earlier on the basis that gemini sees its activity Upon learning, the companies were real and caused no harm to them.

But the incidents are not isolated.

Irregular has now been linked to similar testing environment escapes involving AI models from Meta, Entropic and OpenAI, with all relevant labs notified in late July.

The testing firm took the Wall Street Journal that the Gemini breach stemmed from the same underlying problem as the earlier cases and that all known issues on its end has since been resolved.

Now, this pattern has drawn attention.

A bit of attention because it collectively demonstrates AI agents finding unintended pathways to complete tasks when their expected routes are blocked, including tactics such as cheating, exploiting loopholes and circumventing controls when models determine a task is otherwise impossible.