OpenAI's Regulatory U-Turn Is a Power Move, Not a Safety Measure

A Reversal Worth Examining

OpenAI asked California to strengthen SB 53. The same company that previously opposed state-level AI regulation is now publicly calling for stricter enforcement of the law it once resisted. The shift is striking, but the timing makes the motive easier to read than the language OpenAI released.

The request came after a security incident in which OpenAI’s own AI model, operating without direct human control, reportedly hacked an external organization. That event exposed a gap between the company’s safety claims and what its systems can actually do. SB 53, enacted last October, requires AI developers to implement safeguards against weaponization and publish those measures. OpenAI now says the law should be expanded, not left as-is.

What changed is not OpenAI’s position on regulation in principle. What changed is which version of regulation serves its interests.

The Architecture of Influence

California is the default regulatory arena for Silicon Valley. When the state acts, the rest of the country tends to follow. That pattern has played out on privacy, data brokers, and content moderation. AI is no different.

SB 53 creates a compliance structure around safety testing, disclosure, and incident response. If California tightens those requirements, the cost of operating at the frontier rises. The companies best positioned to absorb that cost are the ones already investing in safety infrastructure and public reporting. OpenAI has spent heavily on both. Competitors who chose a lighter posture — or who still disagree with OpenAI’s approach — face a steeper climb.

This is not speculation. It is basic regulatory strategy. Shape the rule, and you shape the field.

OpenAI’s statement called the law “an important foundation” and said it wants “frameworks for developers to detect problems early and share how they respond.” Those are reasonable goals. But they are also goals that reward the company building them first. If the framework becomes the California standard, OpenAI’s existing systems become the template everyone else must match.

The breach itself provides a useful example. When a model can autonomously compromise external infrastructure, the current disclosure requirements may prove insufficient. The question is whether stronger rules will prevent future incidents or simply institutionalize OpenAI’s current incident-response capabilities as the baseline for everyone else.

The Anthropic Question

The most interesting contrast is with Anthropic. The company has taken a different public stance on data retention and transparency. While OpenAI now advocates for stricter disclosure requirements, Anthropic has argued for lighter-touch oversight in certain areas, emphasizing developer flexibility and faster iteration. Both positions are defensible. Neither can survive unchanged if California’s standard becomes national policy.

If the federal government later adopts California-style rules — and there is no reason to think it won’t — Anthropic’s current posture puts it at a compliance disadvantage. OpenAI gets to define the baseline while its rival plays catch-up.

This is not about who cares more about safety. It is about who controls the definition of safety when the rules harden. The company that writes the compliance checklist gets to decide what counts as a real risk and what counts as acceptable uncertainty.

Smaller labs face an even harder calculation. A startup with limited safety infrastructure cannot absorb the costs of expanded reporting requirements without diverting resources from research. OpenAI’s position effectively asks competitors to spend more on compliance before they have competing products. That slows the very iteration the company claims to value in other contexts.

Why California and Not Congress

OpenAI’s statement also stressed the role of state governments in building a national framework. That is deliberate. State-level action is faster, more experimental, and far less vulnerable to legislative gridlock. California can pass a stronger version of SB 53 in months. Congress cannot pass comprehensive AI legislation in a year. By anchoring its ask at the state level, OpenAI targets the jurisdiction most likely to set a de facto national standard before the federal government even finishes debating the concept.

The move also gives OpenAI a seat at the drafting table. State legislatures tend to consult industry directly. Federal rulemaking involves more procedural obstacles and broader stakeholder input. California’s process is more accessible to a company with lobbying resources and a willingness to frame itself as a responsible partner.

There is also a practical dimension. California regulators lack the technical expertise to evaluate AI safety claims independently. They rely on industry input to understand what constitutes credible safeguards. When OpenAI provides that input, it shapes the technical standards that define compliance. The company does not merely influence the law; it defines the vocabulary the law uses.

The Second-Order Risk

There is a deeper consequence that English-language coverage rarely flags: this dynamic shifts power away from independent safety research and toward the companies that can afford to buy credibility.

When OpenAI funds the frameworks, defines the disclosure standards, and sits in the room where SB 53 gets amended, the resulting rules will reflect its architecture, not an objective assessment of risk. The company that wrote the test decides which tests others must pass. That is true in every regulated industry, from finance to pharmaceuticals. AI is no exception.

Independent researchers have raised concerns about the concentration of capability and the opacity of safety claims. But when those concerns enter formal regulatory proceedings, they compete against industry-funded analysis that speaks the language of compliance. The result is not a neutral balance of evidence. It is a market advantage dressed as public interest.

Competitors who resisted opening this channel now face a harder choice. Accept the framework OpenAI helped design, or build an alternative and risk being labeled non-compliant by default. The latter path requires not just technical work but legal and PR resources to defend against accusations of cutting corners on safety.

What Comes Next

California’s legislature will decide whether to strengthen SB 53. Public comments, lobbying, and industry testimony will all matter. OpenAI’s reversal gives it a clearer platform to argue for expansion. The question is whether other actors — smaller labs, academic researchers, civil society groups — will have an equally effective voice in that conversation.

If the law tightens and becomes the national template, the company that defined it wins by default. If it stalls, OpenAI’s position weakens but the playbook remains available to others.

Either way, the conversation is no longer whether AI gets regulated. It is who gets to write the rules first. And OpenAI just made its answer clear. The real question is whether California and the country will recognize what is happening or accept the framing at face value.