2026-05-04-gpt-55-cyber-restricted-security-model

Summary

OpenAI launched GPT-5.5-Cyber, a GPT-5.4 variant with lowered refusal boundaries for legitimate cybersecurity work. Binary reverse engineering capabilities enabled. Available via Trusted Access for Cyber (TAC) program to government entities, critical infrastructure operators, security vendors, cloud platforms, and financial institutions. AISI evaluation: 71.4% Expert-tier cyber pass rate (Mythos Preview: 68.6%, GPT-5.4: 52.4%, Opus 4.7: 48.6%). Completed AISI’s 32-step “The Last Ones” corporate-network attack range end-to-end in 2 of 10 attempts — second model ever to do so.

Implications

• Direct competitor to Claude Security (public beta since May 1) — both vendors shipping security-specific products the same week
• “Restricted access” model parallels Anthropic’s Enterprise-only Claude Security — security capabilities behind trust gates, not open access
• The Register notes the irony: OpenAI restricts GPT-5.5-Cyber access while criticizing Anthropic for restricting Claude’s capabilities
• Binary reverse engineering is a genuinely new capability for defenders — analyzing compiled software without source code
• Feeds: Claude Code security thread, enterprise deployment thread, agent security surface thread