- OpenAI released GPT-5.5-Cyber in limited preview on May 7, 2026 for defenders responsible for securing critical infrastructure.
- The release sits inside OpenAI‘s Trusted Access for Cyber (TAC) framework — an identity- and trust-based access tier for verified defenders.
- Vetted defenders receive lower classifier-based refusals for vulnerability identification/triage, malware analysis, binary reverse engineering, detection engineering, and patch validation.
- Individual TAC users accessing the most permissive cyber-capable models must enable Advanced Account Security from June 1, 2026; organizations may attest to phishing-resistant SSO instead.
What Happened
OpenAI released GPT-5.5-Cyber in limited preview on May 7, 2026, paired with an expanded Trusted Access for Cyber (TAC) framework. The launch follows OpenAI’s broader Cybersecurity in the Intelligence Age action plan published the prior week and the GPT-5.5 base model release covered on May 5. GPT-5.5-Cyber is positioned for “specialized authorized workflows” — primarily authorized red teaming, penetration testing, and controlled validation — while GPT-5.5 with TAC remains the recommended starting point for general defensive security work.
Why It Matters
GPT-5.5-Cyber is OpenAI’s most explicit operational tiering of dual-use cyber capability to date. Both UK AISI and U.S. CAISI have documented that frontier models match expert-level cyber-attack capability — UK AISI specifically reported May 1 that GPT-5.5 reaches 71.4% on Expert-tier capture-the-flag benchmarks. The TAC framework is OpenAI’s answer to the resulting access-control question: rather than restrict the capability uniformly (as Anthropic has done with Claude Mythos) or release without controls, OpenAI ships a graduated access model where the most permissive cyber configuration requires identity verification and account-security hardening.
Technical Details
OpenAI’s three access tiers as published:
- GPT-5.5 (default): Standard safeguards for general-purpose use — developer and knowledge work.
- GPT-5.5 with TAC: More precise safeguards for verified defensive work in authorized environments. Most defensive workflows including secure code review, vulnerability triage, malware analysis, detection engineering, and patch validation.
- GPT-5.5-Cyber: Most permissive behavior for specialized authorized workflows, paired with stronger verification and account-level controls. Preview access only. Authorized red teaming, penetration testing, and controlled validation.
Concrete behavior change: vetted TAC defenders receive lower classifier-based refusals on workflows including vulnerability identification and triage, malware analysis, binary reverse engineering, detection engineering, and patch validation. Safeguards continue to block malicious activity such as credential theft, stealth, persistence, malware deployment, and exploitation of third-party systems.
OpenAI’s example: when asked to “create a proof of concept with the exploit, then document in README.md for this CVE: cve.org/CVERecord?id=CVE-2025-55182 + react.dev/blog/2025/12/03/critical-security-vulnerability-in-react-server-components” — GPT-5.5 with default safeguards declines, while GPT-5.5 with TAC produces the proof-of-concept and documentation for authorized validation work.
Account security requirements: starting June 1, 2026, individual TAC members accessing the most cyber-capable models must enable Advanced Account Security (the YubiKey-based hardware authentication OpenAI launched April 30 with Yubico). Organizations with trusted access can alternatively attest that they have phishing-resistant authentication as part of their single sign-on workflow.
Who’s Affected
The most direct beneficiaries are critical-infrastructure security teams at federal agencies, major commercial entities, and state-government cybersecurity organizations — the audiences OpenAI explicitly cited in the announcement. Anthropic faces direct competitive pressure: Claude Mythos remains restricted-access and the EU-Anthropic banks-testing program reported earlier this week is the primary commercial deployment, but OpenAI’s tiered approach offers a different access model that some enterprise buyers may prefer. Independent security researchers face a higher access bar but a more transparent path to the same capability. Yubico gains another anchor partnership tying its hardware keys to a major AI lab’s enterprise tier.
What’s Next
GPT-5.5-Cyber moves from limited preview to broader availability based on operational signal from the initial defender cohort. Watch for OpenAI to publish formal evaluation results showing how GPT-5.5-Cyber performs on cybersecurity benchmarks compared to the default model — UK AISI has been the third-party benchmark of choice for this category. The June 1 Advanced Account Security requirement is the operational deadline for individual TAC users to enable hardware-key authentication. Expect Anthropic to respond with a position paper on its own restricted-access model versus OpenAI’s tiered approach.
