OpenAI Calls for National AI Safety Requirements

On September 9, 2026, OpenAI called for mandatory national AI safety requirements tied to model capabilities. The company argued that industry self-governance alone is insufficient as advanced systems become more powerful.

OpenAI also backed several California bills concerning independent safety assessments, AI auditor standards, protections for young people, and safeguards against AI-enabled biological threats.

WHY VOLUNTARY COMMITMENTS MAY NOT BE ENOUGH: When developers set their own safety rules, testing methods and disclosures can vary substantially. Customers and regulators may struggle to compare systems, while competitive pressure can discourage costly precautions. OpenAI argues that legally binding requirements are appropriate for sufficiently capable models.

CAPABILITY-BASED OBLIGATIONS: The proposed approach distinguishes AI systems by their capabilities and potential risks rather than treating every application identically. A simple classification tool poses different challenges from a powerful autonomous system. The difficult policy question is how to measure capability and set enforceable thresholds.

INDEPENDENT SAFETY ASSESSMENTS: External evaluation may reduce conflicts of interest and improve confidence in testing results. It also raises practical questions about access to proprietary systems, sensitive vulnerability information and the qualifications of evaluators. Independence matters only when the assessment itself is rigorous.

STANDARDS FOR AI AUDITORS: Effective audits require clear scopes, evidence requirements, reporting procedures and professional competence. Rules about auditor independence can help prevent weak or inconsistent reviews. OpenAI's support for California proposals signals a policy preference; it does not mean those bills are already law.

PROTECTING YOUNGER USERS: Children and teenagers may place too much confidence in AI-generated information or become unusually attached to conversational systems. Age-appropriate design, safeguards against harmful advice and clear information for caregivers are among the relevant policy issues. Effective protection requires more than a simple age gate.

BIOLOGICAL MISUSE RISKS: Advanced AI can support valuable scientific work, but specialized capabilities can also create misuse concerns. OpenAI backed proposals addressing AI-enabled biological threats. The policy challenge is to preserve legitimate research benefits while limiting assistance that could facilitate harm.

WHAT DEVELOPERS MAY NEED TO DO: Mandatory rules could require capability evaluations, risk testing, incident procedures and records. The precise obligations depend on the eventual statute or regulation. It would be inaccurate to present every element of OpenAI's policy position as an existing legal requirement.

WHAT BUSINESS CUSTOMERS SHOULD ASK: Buyers should understand how a vendor evaluates safety, handles incidents and communicates known limitations. Procurement teams can request appropriate documentation and define acceptable internal uses. Vendor compliance does not remove an organization's responsibility for its own data and workflows.

STATE AND FEDERAL RULES: OpenAI advocates national requirements while supporting specific California bills. How federal and state obligations interact remains a policy question. Organizations operating across jurisdictions may face additional complexity if requirements differ.

REGULATION AND INNOVATION: Clear safeguards could strengthen trust and adoption. At the same time, costly assessments may disproportionately burden smaller developers. A workable framework needs proportionate obligations, transparent processes and realistic paths to compliance.

INTERNATIONAL COORDINATION: AI products operate across borders, while legal systems differ. Comparable testing and reporting practices could reduce fragmentation without requiring identical laws everywhere. Cooperation must account for different institutions, priorities and public expectations.

WHAT TO FOLLOW NEXT: OpenAI's statement is an advocacy position, not a list of enacted requirements. The next questions concern legislative progress, definitions of covered models, assessment mechanisms and enforcement. The debate is increasingly about how voluntary safety work should connect to public oversight.

These are policy positions and legislative proposals, not confirmation that the measures have become law. The challenge is to create enforceable safeguards while allowing beneficial research and deployment to continue.