Three Phone Calls

On Wednesday night, May 21, Elon Musk called the President of the United States. Mark Zuckerberg called next. David Sacks, the administration’s AI czar, followed up Thursday morning. By Thursday afternoon, the White House AI executive order was dead.

Twelve hours. Three phone calls. One dead framework.

The order would have established a voluntary mechanism for AI developers to submit advanced models for security review up to 90 days before public release. No mandatory holds. No licensing regime. Just: show someone what your model can do before you ship it to 300 million people.

That was too much.

Who Made the Calls

Elon Musk runs xAI, which competes directly with OpenAI and Anthropic. He has a structural interest in keeping the regulatory field open because any oversight framework might constrain his own model deployments.

Mark Zuckerberg runs Meta, which has positioned itself as the champion of open-source AI development. Meta’s open-source strategy depends on shipping models without pre-release review. A voluntary submission process, even without mandatory delays, threatened the pace Meta needs to maintain against closed competitors.

David Sacks is the White House’s own AI advisor. He was supposed to represent the public interest in these conversations. Instead, according to reporting from The Washington Post, he helped coordinate the lobbying that killed the order he was charged with implementing.

OpenAI, notably, supported the executive order. So did Anthropic. The companies that build the most capable frontier models wanted oversight. The companies that stood to lose competitive advantage from any delay did not.

What Died With the Order

The executive order was modest by international standards. The EU’s AI Act requires mandatory disclosure, conformity assessment, and sanctions up to 6% of global revenue for non-compliance. Article 50 takes effect August 2, 2026, requiring labeling of all AI-generated content and synthetic interactions across Europe.

The American executive order asked only for voluntary engagement. Still too much.

What this means in practice: the AI models that power your news feed, your recommendation algorithm, your content curation, and your search results will continue shipping without any pre-release review by anyone outside the company that built them. No independent assessment of their behavioral manipulation capabilities. No disclosure of what they optimize for. No requirement to demonstrate they serve the user’s interest rather than the platform’s.

The companies building these systems will continue to self-regulate. And the track record of self-regulation in technology is not encouraging.

The People Who Feed You Information Decided They Don’t Need Rules

This is the part that connects to your morning routine.

The AI that selects which news stories appear in your feed was built by companies whose leadership just demonstrated they will personally intervene at the highest level of government to prevent even voluntary oversight of their models. These are not companies that believe accountability improves their product. These are companies that view accountability as a competitive threat.

Musk controls X, where hundreds of millions of people encounter news content shaped by his platform’s algorithms. Zuckerberg controls Facebook, Instagram, and Threads, which collectively influence the information environment of over three billion monthly users. Between them, they touch a majority of the world’s online news consumption.

And they just killed the gentlest possible version of AI accountability.

The Argument That Won

The pitch to Trump was simple: any AI regulation, however modest, puts America behind China. Musk framed it as a question of national competitiveness. The accelerationist faction within the administration, including staffers in the Vice President’s office, agreed.

This argument has a seductive logic if you accept its premise: that the AI race is zero-sum and that speed matters more than direction. But the premise assumes that the only risk is falling behind. It does not account for the risk of building systems that are fast, capable, and pointed at the wrong target.

An AI model that can outperform China’s best system is not necessarily an AI model that serves American citizens well. Speed and capability say nothing about alignment. And alignment, in the context of news and information systems, means: does this AI optimize for the reader’s understanding, or for the platform’s engagement metrics?

The executive order was the only instrument that might have forced that question into the open. It is gone now.

What Remains

Three layers of protection existed, in theory, between you and an AI system optimized against your interests:

Government regulation. Killed last Wednesday.

Platform self-governance. The same platforms that killed the regulation.

Product-level architecture. How the specific tools you use are built.

Two of those three layers are controlled by the same people. The only layer they cannot reach is the product you choose. The app. The tool. The information source that either extracts your behavioral data or does not. That either optimizes for your engagement or does not. That either subjects its AI to external review or does not.

ntrl’s AI never touches your behavioral data. It reads the article, not the reader. The neutralization pipeline’s only optimization target is removing manipulative language from news content. There is no engagement model to protect from oversight, because there is no engagement model.

When the regulators work for the regulated, the architecture of the product you use is the last line of defense. Choose accordingly.