Forest Global Training Partners

Cyber Risks · The Hacker News

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

Summary: Google announced Gemini 4 Carbon on Wednesday, a frontier model it says leads SEC-bench Pro at 88.2% for finding vulnerabilities, ExploitGym at 54.8% for exploitation, and CWE-bench at 71.0% for patching, and scores 59 on the Artificial Analysis Intelligence Index ahead of Anthropic's Claude Opus 5.5 (58), OpenAI GPT-6 Astra (57), and GPT-6.1 Sol (55). Google says Carbon is twice as likely as Argon to generate secure code, resists every evaluated indirect prompt injection attempt, and is rolling out first to a subset of cyber defenders under its Fairwind Program.

Bottom line: Anthropic disclosed that an unreleased internal model produced a one-click remote code execution in Chrome, a one-click exploit chain in Safari, an Android lock screen bypass, and a DoS flaw in the Windows HTTP.sys driver (CVE-2026-33096, CVSS 7.5) - work Anthropic says was historically the domain of few specialist teams and can now be done with little human involvement.

Read the original at The Hacker News ↗

Topics
AI Security
Mentions
CVE-2026-33096
Related