OpenAI's public policy agenda: safety rules, youth protections

OpenAI says frontier AI safety is "a national security and public safety issue," and it is throwing its weight behind state laws in California, New York, and Illinois — while calling for a federal framework that would eventually preempt those very state rules.

That is the central tension in the company's newly published public policy agenda, a wide-ranging document that lays out OpenAI's positions on model safety, youth protections, education, workforce transition, deepfakes, and the energy costs of AI infrastructure. The agenda arrives as governments worldwide grapple with how to regulate increasingly capable general-purpose models, and as OpenAI itself faces scrutiny over its data center buildout, its handling of user safety, and its relationships with policymakers.

Frontier model safety: states first, then federal preemption

OpenAI frames the most advanced general-purpose AI models as capable of creating risks related to cyber, chemical, biological, radiological, and nuclear (CBRN) weapons. In the US, the company supports state efforts to align around common frameworks — specifically California SB 53, the New York RAISE Act, and Illinois SB 315 — bills that emphasize transparency, public reporting around catastrophic-risk evaluations and safety incidents, whistleblower protections, and enforceable accountability for developers that fail to meet safety and security responsibilities.

"These state-level approaches can help establish harmonized standards that reduce fragmentation and create a path toward an eventual federal framework," the company argues.

That federal framework, in OpenAI's telling, should leverage the emerging consensus reflected in state frontier safety laws and strengthen the Center for AI Standards and Innovation (CAISI) as the US federal government's primary institution for frontier AI safety. OpenAI wants CAISI to conduct evaluations of the most capable frontier models, create an independent assessment ecosystem, and prioritize monitoring progress toward recursive self improvement (RSI) — the scenario in which AI systems improve themselves without human intervention.

The company's position comes with a condition that will draw attention from state regulators: "With a comprehensive federal framework in place, we support the preemption of state laws that seek to regulate the same frontier safety risks."

Internationally, OpenAI notes it was the first US company to sign the EU AI Act Code of Practice, and among the first companies to enter into voluntary agreements with both the US CAISI and the UK AI Security Institute (UK AISI). Looking ahead, the company says policymakers should begin considering "more ambitious ideas" — model containment playbooks, incident reporting, or international governance bodies that facilitate coordination around frontier AI risks, safeguards, and security incidents.

On cybersecurity, OpenAI supports expanding trusted access to AI-powered cyber defense tools, stronger partnerships among governments, researchers, and industry, and modernizing what it calls outdated public-sector cybersecurity systems. The company says it is partnering with the US government at the federal, state, and local levels, and with international partners.

Youth safety and synthetic CSAM

The agenda devotes substantial space to teenagers. OpenAI argues teens should have access to "safe and trustworthy AI" at home, at school, and as they prepare to join the workforce — and supports "strong, enforceable, risk-based regulations" to make that happen.

The specific asks read like a legislative checklist. Privacy-preserving age assurance so companies can identify when a user is a minor. Regular youth safety risk assessments that identify foreseeable risks before harm occurs. Accessible parental controls. Clear public youth safety policies that explain what protections exist and how they evolve.

Crisis-response protocols, OpenAI says, should include in-service support, referrals to real-world resources such as the 988 suicide prevention line in the United States, and timely parental notifications when appropriate. The company also backs restricting targeted advertising to minors and the sale of their personal data, plus independent audits underpinned by common standards that allow interoperable audits across jurisdictions.

OpenAI says it supports the Parents and Kids Safe AI Coalition's efforts to advance these principles.

On child sexual abuse, the agenda calls for modernizing child protection laws to address generative AI's capacity to create synthetic CSAM, digitally alter existing imagery, and scale grooming or exploitation. Policymakers, OpenAI says, should ensure CSAM statutes clearly cover AI-generated and digitally altered material, clarify liability for intentional efforts to produce or solicit CSAM, and preserve strong enforcement authority for prosecutors and law enforcement. The company also backs provider reporting standards meant to improve the quality of CyberTipline reports and help NCMEC and law enforcement act faster.

Education and workforce transition

OpenAI casts educational institutions as critical to building AI literacy, and says educators should remain central to classroom decision-making and set the pace for AI adoption in schools. Its education wishlist includes investments in AI literacy, strong core instruction in subjects like civics, math, science, and computer science, teacher training with protected time for professional learning, and expanded access to AI tools, broadband, and devices for schools, libraries, and community institutions.

On the economy, the company repeats a claim central to its public positioning: "that's why we make ChatGPT available for free." It supports regional AI hubs connecting employers, labor organizations, community colleges, and workforce boards, along with small-business adoption programs providing tools, technical support, and hands-on training.

OpenAI points to its partnerships with labor organizations, including North America's Building Trades Unions, to expand access to training for workers preparing for an economy shaped by increasingly capable AI systems. It also says it supports a more ambitious workforce and economic transition agenda — one that contemplates portable benefits, tax modernization, public wealth funds, and adaptive safety nets.

The company cites its user base as evidence of broad access: as many women as men use its tools, more users both under and over age 30 than any other AI platform, and more users earning under $100,000 than over $100,000 — "mirroring the broader global workforce." OpenAI does not provide underlying figures for these claims in the document.

Deepfakes, provenance, and elections

On synthetic media, OpenAI supports requirements for AI tools to include provenance signals — such as those developed by the Coalition for Content Provenance and Authenticity (C2PA) — in the audiovisual content they generate. It backs protections against harmful misuse of voice and likeness while preserving safeguards for free expression, parody, and journalism.

The agenda calls for criminalizing the distribution of synthetic non-consensual intimate imagery and strengthening remedies for victims. On elections, OpenAI supports prohibitions on intentionally distributing AI-generated content to mislead voters, plus disclosure requirements for certain AI-generated political advertising and campaign communications. The company says it partners with trusted election authorities to provide accurate voting information and to prevent its systems from unduly recommending or ranking candidates.

Infrastructure: paying its own way on energy

Perhaps the most commercially consequential section addresses the physical footprint of AI. Every Stargate site — the joint infrastructure venture building massive data centers — has a tailored community plan developed in consultation with local communities, including on site design and energy sources, according to the document.

"We commit to paying our own way on energy, so that our operations don't increase electricity prices," OpenAI states. The company supports policies requiring datacenters to pay their fair share of electricity costs, including state public utility commissions creating large load tariffs that require datacenters to pay the incremental costs they create.

OpenAI also backs transparency requirements around water, electricity, and government agreements — with exceptions for security, commercial sensitivities, and trade secrets — plus sustainability reporting on technologies like low-emission backup generators and closed-loop or low-water cooling systems. On jobs, the company supports incentives to use registered apprentices and project labor agreements for local construction work.

What it signals

The document is notable less for novelty than for consolidation. OpenAI is aligning itself with the state frontier-safety bills that emerged after California's contested SB 1047 died in 2024, while explicitly conditioning its support on eventual federal preemption — a stance that sets up a fight over whether national standards will raise the floor or cap the ceiling. Its embrace of enforceable developer accountability, independent audits, and whistleblower protections marks a shift from the voluntary-commitments era of AI governance. And its infrastructure positions, including large load tariffs, amount to an attempt to preempt the political backlash building around data center energy costs in the states where Stargate is building. How much of this agenda survives contact with an actual federal bill will determine whether it shapes US AI regulation or simply frames it.