Anthropic, OpenAI, the Vatican and Congress are all trying to answer the same question: who should decide how artificial intelligence is governed? Their answers differ sharply.

Six months ago, I predicted AI would become a political centerpiece in 2026 . That prediction is now taking shape as industry, religious and government leaders publish competing blueprints for AI’s future. What follows are three public arguments about where AI is taking us and one congressional attempt to turn those arguments into law.

Anthropic asks Washington to move faster against catastrophic AI risk. Pope Leo XIV asks society to keep humanity at the center of technological progress. OpenAI asks America to build, compete and lead under clear national rules. The Great American AI Act discussion draft shows Congress testing how those claims might become agencies, audits, disclosures, workforce programs and enforcement.

Anthropic’s Policy On The AI Exponential is the safety realist document. Dario Amodei argues that capability is moving faster than government can comfortably process. He raises practical concerns that frontier models could worsen cyber and biological risks, act autonomously and create loss-of-control scenarios. His answer is direct governance: third-party testing, model security obligations, incident reporting and deployment thresholds tied to catastrophic risk. That safety-first posture has drawn political pushback. The Economist reported that David Sacks, an adviser to the Trump administration on AI, accused Anthropic of running a “sophisticated regulatory-capture strategy based on fearmongering.” The charge shows how contested the AI safety agenda has become.

The Vatican’s Magnifica Humanitas begins from a different premise. AI is judged by whether it protects human dignity, truth, work, freedom, peace and the vulnerable. The encyclical’s focus extends beyond technical and indicates that the danger is a culture that treats people as data, labor as an expendable input and intelligence as the measure of human worth. Jill Lepore wrote in The New Yorker that the encyclical makes a case for “placing moral concerns, and not profit, or competitive advantage, or efficiency, at the center of any discussion of artificial intelligence.”

OpenAI’s 2025 Economic Blueprint and its 2026 policy paper make an industrial and geopolitical argument. The premise is that the United States needs infrastructure, talent, energy capacity and rules to lead the AI era, while also preparing for the labor, tax and governance pressures created by advanced AI. Chris Lehane, OpenAI’s vice president for global affairs, stated on LinkedIn that the common theme in their policy is to “build an AI economy that is open, democratic, and broadly shared by building AI that is free, fair and safe.” OpenAI calls for worker voice in AI deployment, broader AI access, grid expansion, adaptive safety nets, portable benefits, frontier auditing, incident reporting and public input. The dominant message remains speed, scale and access, but the company is now pairing that with a more explicit argument for societal safeguards.

The Great American AI Act is different in kind. It is a legislative discussion draft rather than a manifesto. Its language is built around definitions, duties and agencies. It would require large frontier developers to publish frontier AI frameworks, create transparency reports, use independent verification organizations and report critical safety incidents. It also covers whistleblower protection, AI-enabled fraud, limited preemption for state laws regulating AI model development, labor data, AI education, cybersecurity, open-source security, public datasets and international standards cooperation.

Visions Of AI’s Future Share Guardrails And Split On Priorities

The overlap among these texts is more important than it first appears. All four assume that AI is becoming infrastructure for the economy, national security, education, science and public administration. All four also assume that voluntary norms alone will not be enough. Even OpenAI, the most growth-oriented of the group, argues for federal rules that create trust and predictability.

They also converge on the need for national scale. Anthropic worries that fragmented or slow policy will fail against frontier risk. OpenAI worries that patchwork rules will slow investment and confuse builders. The Great American AI Act addresses this through federal preemption of state laws specifically regulating model development, while preserving state authority over deployment, use, general laws and common-law remedies. The Vatican treats AI as a shared civilizational challenge.

The disagreements begin with what each document protects first, and the four texts pull in different directions from there. Anthropic starts with catastrophic risk, while the Vatican starts with human dignity. OpenAI starts with national opportunity and democratic competitiveness, and the House draft starts with institutional capacity, asking who defines risk, who verifies compliance, who collects labor data and who enforces the rules. Together, these starting points begin to outline the policy field.

Labor is the clearest example, since each document treats the same concern from a different angle. Anthropic gives serious attention to displacement, including the possibility of durable pressure on wages and employment, while the Vatican insists that work is tied to dignity, participation and citizenship. OpenAI emphasizes new jobs, tools and infrastructure, but also acknowledges that AI could disrupt jobs and industries, erode parts of the tax base and require adaptive safety nets, portable benefits and worker participation in deployment decisions.

The Great American AI Act translates the concern into measurement, calling for WARN Act disclosures when AI is a substantial factor in covered mass layoffs, forecasts for AI-sensitive occupations and a study of rapid AI adjustment assistance (the WARN Act is a federal law requiring large employers to give advance notice before mass layoffs or plant closings).

The same pattern appears in security. Anthropic focuses on cyber, biological and loss-of-control scenarios. OpenAI focuses on national leadership and defense capability. The Act turns those concerns into model-weight security, critical safety incident reporting, support for open-source maintainers and research security. The Vatican supplies the moral constraint: safety cannot mean power without accountability, especially when autonomous weapons, surveillance and manipulation are involved.

Visions Of AI’s Future Need Durable Rules

Executive orders can signal priorities, coordinate agencies and set near-term expectations. They are useful instruments, but they are fragile foundations for a technology that will shape investment decisions, labor markets, education systems and public trust for decades. As I argued before , legitimacy comes from accountability rather than pure ambition. The same test applies to the four visions described here. History will judge Anthropic, the Vatican, OpenAI and Congress not by the boldness of their claims but by whether the institutions built to act on them are credible, consistent and accountable.

Durable technology policy matters because it gives the economy a predictable investment scenario. Companies need to know whether federal rules will survive the next election cycle. Workers need confidence that policy will measure displacement honestly and respond with real tools. Citizens need assurance that safety, free speech and privacy rules are not rewritten every four years. Industry also benefits from clarity because uncertainty raises the cost of capital.

The Great American AI Act is valuable because it moves the discussion into that terrain. It may change substantially before introduction and it should. A discussion draft is designed for pressure testing. Rep. Scott Peters, D-Calif., said in the release that the draft “is an encouraging first step toward the bipartisan legislation needed to keep pace with the rapid advancement of AI.” The U.S. Congress is beginning to convert AI principles into a statutory architecture. That is healthy if it invites industry, civil society, labor, academia and state officials into a serious negotiation over acceptable guardrails.

The Dynamic Governance Model I proposed in 2025 offers one template for what that durable architecture could look like. Like the Great American AI Act, it relies on independent audits and third-party verification rather than government fiat to turn standards into something enforceable, and both borrow from existing assurance industries such as product safety certification and financial accounting. But the two differ in reach and theory of accountability. The Act's binding requirements apply narrowly to the largest frontier developers and route enforcement through a single federal office, while the Dynamic Governance Model spreads standard-setting across public-private partnerships and distributes liability across developers, deployers and end users, enforced through Congress, sectoral regulators and the courts together. The narrower path may be easier to pass. The broader path may be more durable once it does.

AI has its boundaries, and this moment is testing AI’s moral limit. The technology can expand what institutions calculate, automate and predict, but it cannot decide what kind of society we want to live in. That judgment remains human, political and moral.

The optimistic reading is that the country is starting to build the right conversation. Public pressure is rising and our policymakers may be moving beyond slogans. Industry leaders are acknowledging that trust is part of the market, not a constraint outside it. If those forces continue to converge, the United States may approach a workable settlement with enough ambition to lead, enough restraint to protect people and enough durable law to turn visions of AI’s future into guardrails the public can understand.