Is Google targeting the AI security market with 'Carbon,' an AI that finds and fixes hacking vulnerabilities? [Global AI Briefing]

[Financial News] Amid growing concerns that artificial intelligence (AI) could be misused as a hacking tool, Google appears to be taking aim at the AI security market with a new frontier model, 'Gemini 4 Carbon,' that finds hacking vulnerabilities and even proposes code to fix them.
In particular, 'Carbon,' which Google unveiled this time, scored 59 on the overall evaluation index (AAII) compiled by global AI model evaluator Artificial Analysis, earning recognition as the top-performing model among global AI models. This has prompted speculation that Google will move in earnest to expand its enterprise security business, leveraging the cyber defense capabilities of its top-performing AI model.
On the 29th (local time), Google DeepMind announced its new AI model, 'Gemini 4 Carbon.'

'Carbon' is a model designed to handle complex software development and tasks that run for extended periods. In particular, it supports multiple stages of cyber defense, from detecting security vulnerabilities and verifying whether they can be exploited to writing code to fix them.
According to evaluation results released by Google, Carbon scored 88.2% on SEC-bench Pro, which tests its ability to find vulnerabilities in real-world software; 54.8% on ExploitGym, which evaluates whether vulnerabilities can be exploited; and 71.0% on CWE-bench, which assesses its ability to fix security flaws.
Google emphasized, “Internal evaluations show that the success rate for generating secure code has at least doubled compared with our previous model, 'Gemini 4 Argon.'” It added, “The model can work for days to find the root cause of errors in large codebases and fix them.”
This could reduce the time and cost security experts spend analyzing the causes of vulnerabilities and developing fixes after discovering them.
Google decided not to release Carbon to all users immediately, but to provide it first to a select group of cyber defense experts.
Google is also participating in the U.S. government's pre-release model safety evaluation process and plans to gradually expand access, starting with vetted partners.
As Google unveils its new frontier model, 'Carbon,' and highlights its security features, speculation is mounting that the company will make a full-scale push into the enterprise cybersecurity market using AI. Automating vulnerability detection and remediation with AI could speed up companies' security responses and reduce operating costs.
[email protected] Lee Gu-sun Reporter