Preparedness Framework Risk Rating: Medium

What it is

The card discloses that GPT-5 received a 'medium' risk classification under OpenAI's internal Preparedness Framework across four categories: CBRN, cybersecurity, persuasion, and model autonomy, and that this rating governs the conditions under which deployment was authorized.

Why it matters (compliance & governance perspective)

This provision establishes the safety classification baseline that governs GPT-5's deployment conditions, and the 'medium' rating across CBRN and cyberoffense categories may trigger due diligence obligations for enterprise deployers in regulated sectors or those subject to dual-use research oversight frameworks.

Consumer impact (what this means for users)

The document establishes that GPT-5 operates under mitigations corresponding to a 'medium' Preparedness Framework rating, meaning certain high-risk output categories are subject to trained refusal behaviors and monitoring controls that apply to all users regardless of access tier.

Institutional analysis (Compliance & governance intelligence)

1) REGULATORY LANDSCAPE: The 'medium' CBRN and cyberoffense ratings disclosed in this card may interact with U.S. dual-use research of concern policies, Export Administration Regulations administered by the Department of Commerce, and the EU AI Act's systemic risk provisions applicable to general-purpose AI models above defined capability thresholds. The FTC retains authority to evaluate whether safety rating representations constitute unfair or deceptive practices if the disclosed mitigations are found insufficient in practice. 2) GOVERNANCE EXPOSURE: High. The disclosed 'medium' rating across CBRN and cybersecurity categories indicates residual capability risk that has not been fully mitigated, and downstream operators integrating GPT-5 into products may inherit disclosure and risk management obligations depending on their sector and jurisdiction. The card does not specify whether the 'medium' rating reflects pre-mitigation or post-mitigation capability levels, which creates interpretive uncertainty for compliance teams. 3) JURISDICTION FLAGS: EU deployers face heightened exposure under the EU AI Act's GPAI systemic risk provisions. U.S. deployers in defense-adjacent, healthcare, or critical infrastructure sectors should assess whether the CBRN capability disclosures trigger internal review obligations. Illinois, California, and New York have enacted or are developing AI transparency requirements that may require additional disclosure. 4) CONTRACT AND VENDOR IMPLICATIONS: Enterprise procurement teams integrating GPT-5 should assess whether vendor agreements with OpenAI adequately address liability allocation in the event that residual CBRN or cyberoffense capabilities are elicited through their deployment configurations. The card does not assert OpenAI liability for downstream misuse enabled by operator misconfiguration. 5) COMPLIANCE CONSIDERATIONS: Compliance teams should document the Preparedness Framework rating as part of their AI risk inventory and assess whether their use case interacts with any of the four rated risk categories. Organizations in regulated sectors should evaluate whether additional contractual protections or operational controls are required beyond those disclosed in the system card.

Applicable agencies

Provision details

Other risks in this policy

• Unified Model Routing Architecture medium
• Operator-Level System Prompt Controls medium
• External Red-Teaming and Third-Party Safety Evaluation low
• Refusal Training and Output Mitigation Measures medium
• CBRN Capability Disclosure high
• Model Autonomy Risk Category high