OpenAI’s Commercial Cyber‑Attack AI: Pricing, Capabilities, and Global Security Impact

OpenAI has expanded its Daybreak cybersecurity partner program by introducing two new access tiers: Daybreak Blue for authorized defensive teams and Daybreak Red for authorized offensive teams. Daybreak Blue is priced at $30 per million tokens, while Daybreak Red’s GPT‑5.6‑Cyber model costs $75 per million tokens, a 2.5× premium.

How is offensive capability priced?

The higher price reflects OpenAI’s calculation of the market value of a tool capable of complex dual‑use tasks, raising the question of whether the premium captures actual utility or the responsibility and risk associated with weaponizable AI.

What GPT‑5.6‑Cyber can do

Released on 10 August 2026, GPT‑5.6‑Cyber is the flagship model of the Daybreak Red layer, engineered for offensive security tasks such as vulnerability research, exploit‑chain development, authentication bypass, and privilege escalation. In contrast, Daybreak Blue is positioned as a defensive, general‑purpose model that remains subject to security filters and is optimized for multi‑step red‑team workflows.

Reported performance: 95% task‑completion?

OpenAI reports that GPT‑5.6‑Cyber achieves a 95% completion rate on advanced security tasks, markedly higher than the 57.3% reported for its predecessor GPT‑5.5‑Cyber. The company also claims the model has identified over 400 kernel privilege‑escalation vulnerabilities, including a V8 sandbox‑escape (CVE‑2026‑15903) and several undisclosed bugs in a mobile OS, a database, and a popular operating‑system kernel. These figures have not been independently verified.

Who can access the models

Access is limited to a vetted list of service and technology partners. Service partners include Accenture, IBM, Capgemini, EY, KPMG, PwC, NCC Group, and SpecterOps. Technology partners include Unit 42, CrowdStrike, Cisco, Sophos, Akamai, Fortinet, and Cloudflare. Starting 1 September 2026, all personal Daybreak accounts must use a hardware security key.

Cyber‑security AI competitive landscape

The market for security‑focused AI is fragmenting along safety and commercialization lines. Microsoft employs MAI‑Cyber‑1‑Flash (CyberGym score 96%) in its Project Perception platform, while Google restricts Gemini 3.5 Flash Cyber to government use. Anthropic’s Mythos project remains paused due to U.S. export controls issued in June 2026. OpenAI positions Daybreak as the most broadly commercialized offering, with competitors either locked to enterprise, limited by government policy, or halted entirely.

The inherent paradox of the product

The core challenge lies in the dual‑use nature of these models. OpenAI has warned that its Astra model could reach a threshold where it autonomously creates zero‑day exploits and designs end‑to‑end attacks. When a model can identify hundreds of privilege‑escalation bugs, the line between defensive research and offensive weaponization blurs. Partner‑based access restrictions are the current mitigation, but the sustainability of this approach as models become more autonomous remains an open industry question.

Reference: bleepingcomputer.com