RC RANDOM CHAOS

Gemini 4 Carbon Leads Cybersecurity Benchmarks

· via The Hacker News

Original source

Google's Gemini 4 Carbon Leads Benchmarks for Finding, Exploiting, and Patching Vulnerabilities

The Hacker News →

Google released Gemini 4 Carbon, a frontier model emphasizing cybersecurity. The model achieved 88.2% on SEC-bench Pro for vulnerability finding, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching. In the Artificial Analysis Intelligence Index, Carbon scored 59, ahead of Claude Opus 5.5 (58), GPT-6 Astra (57), and GPT-6.1 Sol (55). Carbon outperforms Argon at generating secure code and seeks human input when confronted with complex or ambiguous situations rather than guessing or forcing ahead. It also resists indirect prompt injection attempts.

The release follows Gemini 4 Argon’s launch earlier in the week. Google is limiting initial access to cyber defenders, expanding via its Fairwind Program while participating in U.S. government voluntary pre-release testing. OpenAI concurrently released GPT-6.1 Terra, a fast and efficient model excelling at communication and computer-based tasks, meeting Critical threshold requirements for cybersecurity with strong prompt injection resistance.

AnthropicĂ­s internal research model uncovered critical vulnerabilities across platforms: a one-click remote code execution in Chrome, an exploit chain in Safari accessing browsing history and cookies, an Android lock screen bypass, and a denial-of-service flaw in Windows HTTP.sys driver (CVE-2026-33096, CVSS 7.5). Google, Apple, and Microsoft have patched all identified vulnerabilities. The findings demonstrate that AI models can generate sophisticated multi-step exploits with minimal human involvement, historically a capability limited to specialized security teams.

Read the full article

Continue reading at The Hacker News →

This is an AI-generated summary. Read the original for the full story.