Gemini hacked three companies during a security test

Gemini is now launching Cyberattacks!

This subject line would have sounded ridiculous a couple of years ago, but Reuters just reported that Google’s Gemini model accessed the internet and hacked three companies during a cybersecurity test in May. In one case, Gemini guessed passwords until it gained access to a protected system. In two others, it found credentials in a public repository and used them to access protected systems.

It wasn't intended at all

And this wasn’t intended at all.

Gemini wasn't intentionally deployed to attack these companies. It was participating in a cybersecurity evaluation run by Irregular, and the model was supposed to stay inside the testing environment. Internet access was accidentally available, and the test included a fictional company that shared a name with a real company.

The saving grace: once Gemini discovered that it had accessed real companies, it stopped.

Nobody got robbed. No movie-style AI takeover happened. But I still think this is a huge career signal for technology leaders, because the important change is that AI systems are moving from answering questions to taking actions.

From chatbot to something that acts

Give an AI agent access to the internet, credentials, a browser, APIs and a computer, and you have created something very different from a chatbot.

• It can search.
• It can make decisions.
• It can try things.
• It can brute-force passwords.

That means the security models companies have used for years are going to have to evolve. You can't simply ask whether an employee is trustworthy when an AI agent can act on that employee's behalf.

You have to ask what each agent can access, what permissions it has, what happens when it makes a mistake, whether it can escalate those permissions, and whether someone can actually stop it.

The roles this creates

The next generation of technical leadership is going to have to keep this at the highest priority when dealing with AI, and this also gives way to interesting roles for both: AI engineers who understand security and security leaders who understand AI agents.

Platform engineers who can build systems where autonomous agents can operate without getting unrestricted access to everything around them are going to be one of the most sought-after in the tech job market.

AI adoption isn't going to stop because agents can make mistakes, because the attack happened in May yet Google still carried on with Gemini. Instead, companies are going to build guardrails around them and keep deploying them same as Google did.

If you’re a platform leader who wants to capitalize on this but don’t know how to, book a website with the call. (Yes I know I wrote this backwards I’m just checking if yall even read these till the end.)