This page may contain stale information. Last updated: 2026-06-01

Definition

The Trusted Access for Cyber (TAC) program is OpenAI’s controlled access framework for cybersecurity professionals to use specialized AI models for authorized security research and defensive operations. It was launched alongside GPT-5.5-Cyber on May 7, 2026.

Contradiction

openai and this page cite GPT-5.5-Cyber/TAC launch on 2026-05-07, while pipeline source 2026-05-08-openai-gpt-55-cyber-release uses slug date 2026-05-08. Verify exact announcement date against OpenAI official release.

Overview

TAC addresses the dual-use nature of advanced AI cybersecurity capabilities by implementing tiered verification and strict safeguards. The program is specifically designed to empower “defenders” - government agencies, critical infrastructure providers, and enterprise security teams.

Program Structure

Tiered Verification

Access levels are granted based on:

  • Verified identity
  • Organizational affiliation
  • Use case justification

Higher tiers unlock more advanced capabilities:

  • Basic vulnerability research
  • Penetration testing assistance
  • Red-teaming exercises
  • Exploit development for authorized assessments

Partner Organizations

Initial TAC program partners include:

  • Cisco
  • Intel
  • Snyk

Individual security professionals can verify via chatgpt.com/cyber.

Strategic Context

OpenAI’s TAC program is viewed as a direct competitive response to Anthropic’s Claude Mythos and Project Glasswing initiative. Together, they represent the “AI cybersecurity arms race” in 2026.

Five-Pillar Action Plan

In late April 2026, OpenAI unveiled a strategy to “democratize AI-powered cyber defense”:

  1. Democratizing Defense: Scale TAC to support thousands of defenders globally
  2. Coordination: Collaborate with governments and industry on threat models
  3. Capability Strengthening: Improve model performance for defensive tasks
  4. Visibility and Control: Monitor for potential misuse, maintain “post-launch levers”
  5. User Empowerment: Integrate security features into ChatGPT for general users

Sources