Google announced Gemini 4 Carbon, a frontier AI model featuring specialized cybersecurity and long-horizon capabilities.
Carbon achieved 88.2% in SEC-bench Pro and includes safeguards designed to resist indirect prompt injection while assisting developers and defenders.
Google said the model scored highly on specific benchmarks for vulnerability detection, attack generation, and code remediation. The Gemini 4 Carbon is designed to ask for human assistance when faced with ambiguities rather than trying to guess the solution, and it also has increased resistance to command-line manipulation attempts.
The new model is initially available to a limited group of security experts and partners as part of the company's testing programs. Its release comes just days after the introduction of the previous model in the series, the Gemini 4 Argon.
In the same field, OpenAI presented the new GPT-6.1 Terra model, which features improved communication and security capabilities, while Anthropic recently revealed that an internal research model managed to identify serious problems in widely used programs and systems, which have already been fixed by the respective companies.
G
HACKS.GR / GOOGLE SEARCH
Hacks.gr on Google
Set Hacks.gr as a preferred source.
Preferred source
↗
end of article
Top comments (0)