New12,480+ smartphones indexed — free sandbox tier on every plan.Edge-cached JSON responses, typically under 100ms.SDKs for C#, TypeScript and Python, plus an MCP server for AI agents.New12,480+ smartphones indexed — free sandbox tier on every plan.Edge-cached JSON responses, typically under 100ms.SDKs for C#, TypeScript and Python, plus an MCP server for AI agents.
Get Free API Key
Artificial Intelligence & Cybersecurity•4 min read•2026-11-06

OpenAI Launches GPT-6 Cyber Frontier Model for Autonomous Vulnerability Defense

OpenAI launches GPT-6 Cyber, a frontier AI model rated Critical for cybersecurity that uncovered 81 Firefox zero-days and 948 Linux kernel exploits with automated patch synthesis.

Tech News Bot
DeviceSpecs Intelligence
Original Source Reference
#OpenAI#GPT-6 Cyber#Cybersecurity#AI Agents#Vulnerability Research#Linux Kernel#ExploitGym#Daybreak Red

OpenAI has officially launched GPT-6 Cyber, an advanced frontier model fine-tuned specifically for defensive vulnerability discovery, exploitability verification, attack surface modeling, and automated remediation. Marking OpenAI first Critical rating under its Preparedness Framework, the model uncovered 81 novel Firefox vulnerabilities and generated working exploits for 948 Linux kernel flaws during pre-release testing alongside candidate patches.

Frontier Cybersecurity Benchmarks and ExploitGym Results

Built upon OpenAI GPT-6 Astra architecture, GPT-6 Cyber achieves a 48.5% score on the ExploitGym benchmark—a 51% relative leap over GPT-5.6 Cyber 32.1% baseline. In remote exploit assessments, the model resolved 95% of tasks on RemoteExploitBench. OpenAI notes that while earlier models acted reactively on code audits, GPT-6 Cyber proactively explores complex execution paths, generates reproducible proof-of-concept tests, and validates whether a reported bug is practically exploitable in production environments.

Real-World Demonstrations Across Firefox and the Linux Kernel

In pre-deployment trials conducted with open-source communities, GPT-6 Cyber identified and proved 81 novel high-severity vulnerabilities in Mozilla Firefox and demonstrated that 30 previously known issues carried higher severity than initially cataloged. When applied to the Linux kernel codebase, the agent reproduced 1,724 potential vulnerabilities and developed working exploits for 948 of them—over five times the total achieved by GPT-5.6 Sol. OpenAI coordinated responsible disclosure with Mozilla and submitted candidate patches for 304 critical Linux vulnerabilities directly to subsystem maintainers.

Daybreak Red Deployment Controls, Safety Architecture, and Pricing

Because GPT-6 Cyber achieves a Critical rating for cyber capability under the Preparedness Framework, OpenAI is gating access strictly through its vetted Daybreak Red program for approved enterprise security teams, critical infrastructure operators, and defensive research partners. In scope-boundary evals, GPT-6 Cyber demonstrated 99.93% compliance without guard prompts and 100% containment with prompts. The model features a 1.05-million-token context window and is priced at $20.00 per million input tokens and $100.00 per million output tokens, backed by private safety processing and an ongoing $1 billion defender subsidy initiative.