On November 5 OpenAI launched GPT-6 Cyber for authorized security research: ExploitGym rose from GPT-5.6 Cyber's 32.1% to 48.5% (about 51% relative), out-of-scope exploitation fell to 0.07% without deployment prompts and 0% with them. The model helped find 81 novel high-severity Firefox bugs and develop exploits for 948 Linux privilege-escalation cases, and OpenAI raised its Preparedness Framework cyber rating from High to Critical. It is initially available via API and Codex to select Daybreak Red customers, with expansion to all Red customers planned.
Key Takeaways
- ✓ExploitGym 48.5% vs GPT-5.6 Cyber 32.1% (~51% relative gain; vendor figures)
- ✓Scope alignment: 99.93% avoided out-of-scope services without deployment prompts; 100% with them
- ✓Real-world: 81 novel high-severity Firefox vulns; 1,724 Linux vulns reproduced, 948 exploits developed
- ✓Preparedness raised to Critical; RemoteExploitBench 95% completion (Astra refused 100%)
- ✓Select Daybreak Red via API/Codex first; Daybreak Blue alpha for Astra defensive use expanding
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Core Background & Industry Pain Points
Authorized red teams need models that find, validate, and help patch vulnerabilities without leaving scope. GPT-5.6 Cyber still left gaps. OpenAI's November 4 Linux pipeline hinted at a stronger cyber model; on November 5 it launched GPT-6 Cyber.
Architecture Highlights & Internals
Targets authorized security workflows: understanding software, finding vulns, developing working exploits, plus scope-alignment training and misuse detection such as Private Safety Processing. Access is via Daybreak Red. OpenAI says it beats GPT-6 Astra on most cyber tasks.
Authoritative Benchmarks & Measured Scores
Vendor figures: ExploitGym 48.5% (5.6 Cyber 32.1%); web pentest recall 40.3% (Astra 32.0%); RemoteExploitBench 95% (Astra refused 100%); out-of-scope avoidance 99.93%/100%. Field: 81 novel high-severity Firefox issues and 948 Linux privilege-escalation exploits. No independent third-party replication yet.
Developer Hands-on Guide
Select Daybreak Red customers get API and Codex access now, with expansion to all Red and GPT-5.6 Cyber for all Red. Apply via Daybreak; PSP and IP allowlisting matter for teams with existing red-team process.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.