OpenAI has formally launched the total model of GPT‑5.5‑Cyber, a specialised AI mannequin engineered for superior vulnerability detection, patch technology, and automatic remediation at machine pace.
The discharge is a part of OpenAI’s broader Dawn initiative, which goals to democratize defensive cybersecurity capabilities for trusted organizations worldwide.
GPT‑5.5‑Cyber delivers state-of-the-art outcomes throughout three main cybersecurity analysis benchmarks:
- CyberGym: 85.6% (vs. 81.8% for GPT‑5.5), the best single-model rating recorded.
- ExploitGym: 39.5% (vs. 25.95% for GPT‑5.5), testing exploit technology from identified vulnerabilities.
- SEC-bench Professional: 69.8% (vs. 63.1% for GPT‑5.5), evaluating long-horizon vulnerability discovery throughout advanced software program targets.
The mannequin can navigate massive codebases, hint assault paths, validate exploitability, generate focused patches, and produce remediation proof all inside a single automated workflow.
Codex Safety Plugin Up to date
Alongside the mannequin launch, OpenAI has up to date the Codex Safety plugin, now able to deep codebase scanning with automated patch technology. Since launching in analysis preview in March 2026, Codex Safety has:
- Scanned over 30 million commits throughout greater than 30,000 codebases
- Processed over 70,000 manually verified fixes
- Robotically resolved over 500,000 findings
The plugin integrates straight into developer workflows, supporting SARIF exports, CodeQL queries, and current vulnerability administration pipelines. It generates severity-rated stories with affected code areas, assault path tracing, and codebase-specific patches for human evaluate.
OpenAI launched Patch the Planet, a collaborative initiative co-founded with Path of Bits and partnered with HackerOne and Calif, to handle the crucial vulnerability remediation hole in open-source software program. Greater than 30 open-source tasks have dedicated to take part, together with:
- cURL, Go, Python, Sigstore, and pyca/cryptography
An preliminary five-day dash throughout a number of tasks surfaced tons of of points, merged dozens of patches, and constructed reusable fuzzing and variant-analysis workflows. Collaborating tasks obtain ChatGPT Professional, conditional Codex Safety entry, and API credit.
GPT‑5.5‑Cyber is distributed completely by means of a restricted launch to verified, trusted defenders. It isn’t accessible for normal use. OpenAI has confirmed Trusted Entry for Cyber partnerships with Australia, Canada, France, Germany, Japan, South Korea, and EU establishments, together with ENISA.
OpenAI coordinated pre-deployment testing with the Middle for AI Requirements and Innovation (CAISI) and labored with the Workplace of the Nationwide Cyber Director (ONCD) on the implementation of the June 2026 Govt Order on AI safety.
For many organizations, GPT‑5.5 with Trusted Entry for Cyber and Codex Safety stays the beneficial entry level, with GPT‑5.5‑Cyber reserved for defenders requiring the best functionality tier with enhanced monitoring and scoped controls.
OpenAI’s announcement alerts a basic shift within the cybersecurity risk mannequin. The historic bottleneck of discovering vulnerabilities has given strategy to a brand new problem: patching them at scale.
With Dawn unifying frontier AI fashions, Codex Safety workflows, open-source partnerships, and important infrastructure collaboration, OpenAI is positioning AI-driven remediation, not simply detection, as the following frontier in cyber protection.
What Options Ought to AI SOC Have? – Obtain Free 2026 AI SOC Options Guidelines
