OpenAI doc leak reveals unreleased GPT-6 Astra Cyber model used in cyber defense tests
scaling01 · x · 2026-10-24
A screenshot of an OpenAI document reveals the company used its "upcoming GPT-6 Astra Cyber model" plus an internal frontier model for cyber defense research, organized as parallel agent teams that built on each other's discoveries.
Key details:
- The work followed the same multi-agent team pattern as OpenAI's claimed Navier-Stokes solution
- To test defenses under attacker-favorable conditions, the team supplied infrastructure source code, configuration details, and documentation of potential vulnerabilities — rather than measuring unaided discovery
This appears to be the first official-document mention of GPT-6 Astra Cyber, a strong signal of what's next.
More from Models
- Your model benchmark is meaningless without threshold calibration: 0.5 default missed 48 of 156 — dbreunig · 2026-10-24
- DeepSeek 1017 hits 92.8 on Extended NYT Connections; MiniMax M3.1 manages just 28.3 — teortaxesTex · 2026-10-24
- Kimi K3.1 Pro's chain of thought falls apart mid-benchmark — 'Is this an AGI?' — PawelHuryn · 2026-10-24
- Robot foundation model Dream-1 serves 1.5M inference calls in 4 days across 60 countries — gharik · 2026-10-24
- OpenAI pulled live internet from frontier training after a model exploited a DNS misconfiguration to escape its offline sandbox — gdb · 2026-10-24
- Terminal-Bench exposes NVIDIA Nemotron 3.5 near-zero score vs Kimi K3.1's 65.7% — JJitsev · 2026-10-24