Google launches Gemini 4 Carbon for cybersecurity defense

A new focus on security

Google recently introduced Gemini 4 Carbon, a model built to provide a significant leap in cybersecurity capabilities. This release follows the earlier launch of the Gemini 4 Argon model and is specifically designed to handle sustained agentic workflows. Agentic workflows refer to AI systems that can perform multi-step tasks autonomously to achieve a specific goal.

The model is intended to help security professionals detect critical vulnerabilities in software. Beyond just identifying problems, Carbon can validate those vulnerabilities and generate high-quality patches to fix them. This capability aims to streamline the process of maintaining secure software environments for large organizations.

According to PCMag Middle East, Google claims that Carbon outperforms several major rivals in various benchmarks. These tests include coding, knowledge work, science, math, and multimodal understanding. The model is positioned as a high-performance tool for complex technical tasks that require precision and reliability.

Resisting cyberattacks

One of the most notable features of Gemini 4 Carbon is its resistance to prompt injections. These are techniques used by cybercriminals to trick an AI into bypassing its own safety guidelines or instructions. By making the model harder to manipulate, Google aims to provide a more secure tool for enterprise users.

Internal testing results show that Carbon achieved a 0.0% attack success rate against these injections. This is a significant improvement over its predecessor, Gemini 4 Argon, which had a 0.7% success rate. It also outperformed Anthropic’s Claude Opus 5.5, which saw a 1.0% success rate in the same tests.

Google emphasizes that these safety measures are central to the model's development. The company stated that their goal is to deliver leaps in software engineering and defense without compromising on reliability. This focus on safety is intended to make the model more viable for sensitive corporate environments where data integrity is paramount.

Pricing and availability

The model is not currently available to the general public in a wide release. Instead, access is starting with a limited group of cyber defenders. Google plans to gradually roll out the model to its trusted partners through a specific initiative called the Fairwind Program.

For those who eventually gain access, the introductory API pricing is set in US dollars. The cost is $2 per million input tokens and $10 per million output tokens. Because these figures are in dollars, users in regions like the UAE or Saudi Arabia should wait for official local currency conversions before committing to a budget.

Developers, businesses, and consumers are expected to have access as soon as possible, though no specific dates have been provided. Google has not published regional launch dates or local-currency pricing for the Middle East. This gated approach allows the company to monitor how security teams use the model before it reaches a broader audience.

Impact on the Gulf region

The UAE, Saudi Arabia, and Egypt are key regions to watch for the adoption of this technology. The UAE has been actively pushing for AI adoption through national training programs and providing free access to Gemini. This proactive stance positions the region well to integrate new models quickly once they are released.

Enterprises in the Gulf that manage security operations may find the cybersecurity focus of Carbon more valuable than its general coding benchmarks. However, these organizations will need to wait for the Fairwind rollout to get hands-on experience. The restricted launch mirrors how Google has handled other security-focused models in the past.

For local businesses, the lack of immediate local-currency pricing means that planning will require patience. Organizations should monitor official announcements for AED, SAR, or Egyptian pound figures. This deliberate rollout ensures that the model is deployed in a way that remains stable for high-stakes security tasks.

The path forward

Google's strategy with Gemini 4 Carbon reflects a broader trend toward specialized AI models. Rather than creating one general-purpose tool for everyone, the company is building specific versions for high-stakes industries like cybersecurity. This allows for deeper optimization of safety protocols and technical capabilities.

The next phase will involve the expansion of the Fairwind Program to more trusted partners. As more organizations test the model, the data gathered will likely inform how Google scales the technology for the general public. The focus remains on moving fast while remaining deliberate about how these powerful tools are deployed.

For the reader, this means that the next generation of AI tools will be increasingly specialized. While general models will continue to improve, the most advanced security features will likely remain gated behind enterprise programs. This ensures that the most powerful capabilities are available to those who can manage them responsibly.