中文翻译
摘要: OpenAI Launches GPT-5.5-Cyber for Automated Vulnerability Detection and Patching
Lucas Martin
June 23, 2026
Categories:
Cyber Security News
OpenAI has officially launched the full version of ...
正文
OpenAI Launches GPT-5.5-Cyber for Automated Vulnerability Detection and Patching
Lucas Martin
June 23, 2026
Categories:
Cyber Security News
OpenAI has officially launched the full version of GPT‑5.5‑Cyber, a specialized AI model designed for advanced offensive and defensive cybersecurity workflows.
Released under the company’s Daybreak initiative, the model marks a significant step toward AI-driven vulnerability discovery and automated patch generation at machine speed.
GPT‑5.5‑Cyber sets a new record on CyberGym, a benchmark measuring whether an AI agent can reproduce known vulnerabilities in controlled software environments.
OpenAI Launches GPT-5.5-Cyber
The model achieved 85.6%, surpassing its predecessor, GPT‑5.5, at 81.8%, and outpacing competing models, including Mythos 5 (83.8%) and Claude Opus 4 (73.1%).
GPT-5.5-Cyber Tops CyberGym Benchmark (Source: OpenAI)
The model also demonstrated superior performance on two real-world security benchmarks:
ExploitGym: 39.5% vs. 25.95% for GPT‑5.5 — evaluating the ability to convert known vulnerabilities into working exploits, achieving unauthorized code execution
SEC-bench Pro: 69.8% vs. 63.1% for GPT‑5.5 — measuring long-horizon vulnerability discovery and proof-of-concept generation across complex software targets
Access remains restricted to verified defenders through a continued limited release, with stronger verification, scoped controls, and requirements for human oversight.
Alongside the model release,
OpenAI updated the Codex Security plugin
, which has already scanned over 30 million commits across more than 30,000 codebases since its March 2026 research preview.
Human reviewers have manually marked over 70,000 findings as fixed, and more than 500,000 additional findings have been automatically resolved.
The updated plugin enables out-of-the-box defensive security workflows by integrating directly into developer environments.
It generates severity-ranked findings with validation evidenc
采集时间: 2026-06-23 20:16:32
AIGPTOpenAIClaude人工智能