Back to the stories

Google announces Gemini 4 Carbon frontier model for cyber defenders

Score 10.0,

Artificial Intelligence

The next step for AI in cybersecurity is not just finding weak spots. It is helping defenders test them and write repairs.

Google says its new Gemini 4 Carbon can search real software for flaws, check whether they can be exploited, and help produce fixes. Google describes it as its strongest model.

For now, Carbon is available only to a subset of cyber defenders. Google plans to expand access gradually to trusted partners, but has not given a public release date.

That matters because spotting a flaw is only the start. Defenders also need to know whether it can be used to cause harm, then work out how to fix it. The ambition is striking: one system could help move through all those steps.

In Google’s internal tests, Carbon generated secure code successfully more than twice as often as Gemini 4 Argon. Think of it like a security team that finds a weak lock, tests whether it actually opens, then drafts a repair.

This is a controlled rollout, not a tool every business can use today. Google says it plans broader access through its products, but the timing is unknown.

The real test is whether Carbon can find and help fix useful flaws in messy, real-world software. How much of that work can defenders safely hand over?