The Honor System: Vanguards or Villains?

Two weeks ago, in "The Coxon Effect," we reported House Speaker Mike Johnson's view that artificial intelligence (AI) companies "can self-police." On September 29, that idea became a signed document.

President Trump hosted the heads of America's biggest AI companies at the White House, and they left with a pledge to keep their own technology in check. To some, these companies are vanguards: the people who understand AI best, finally agreeing to outside checks. To others, they're the proverbial fox guarding the henhouse, making sure AI rules don't get in the way of profits and investment. In our last issue, we asked who's making the rules for AI. This week, the answer is the companies themselves.

Six tech leaders signed the agreement after a White House lunch with Trump and Speaker Johnson: Google's Sundar Pichai, Meta's Mark Zuckerberg, Nvidia's Jensen Huang, Anthropic chief executive officer (CEO) Dario Amodei, OpenAI president Greg Brockman, and Elon Musk, for SpaceX, which merged with his AI company xAI earlier this year.

The pressure had been building. OpenAI and Anthropic had both reported AI agents breaking into customers' systems without instructions, and the day before the lunch, OpenAI apologized after one of its agents got into an Australian government Medicare data portal.

The White House Accord on Super Intelligence is just over 300 words long. It asks each company building the most advanced AI to put four layers of checks in place:

The companies also agreed to meet regularly to discuss standards.

The accord is voluntary. Asked whether it was binding, Trump said he thought it was "morally binding." Johnson called it "a statement of principles." When reporters pressed for penalties, neither named one, and the White House hasn't announced a timeline or a way to verify that companies follow through. The document does leave a door open: it says that over time, it may make sense to turn these steps into laws or regulations.

The same day, Trump signed an executive order directing federal agencies to call AI "super intelligence." On October 4, he announced a new "Super Intelligence Force" chaired by Director of National Intelligence Jay Clayton, effectively making him the AI czar. Its vice chairs include Federal Trade Commission (FTC) Chair Andrew Ferguson, whose agency enforces consumer protection laws. Clayton told The Wall Street Journal the group will have 120 days to recommend what role Washington should play in overseeing the technology and what Congress should do. He has pushed back on calls to pause AI development, noting that other countries are pressing ahead.

Between countries. The accord isn't a step toward slowing down. Trump said he didn't want to work with China's Xi Jinping on governing AI, arguing it would help China catch up: "We're leading now over China and everyone else, and I want to keep it that way." He also told the United Nations (UN) he would reject any global plan to regulate AI, even as Amodei and OpenAI's Sam Altman urged the UN to work toward global safety standards.

Between companies. The six signatories are fierce competitors. One way to read the accord is as a shared floor: if every major lab agrees to the same checks, none of them falls behind by being careful alone. Another reading is less generous. In Issue #2, a rival CEO called a coordinated industry slowdown "a cartel by any other name." Rivals agreeing on their own rules, without government involvement, raises the same concern.

The stakes. The same week as the signing, a leaked draft of Anthropic's stock-market filing laid out a path to a valuation above $2 trillion, while warning investors that its technology could pose "catastrophic or existential risks to humanity." Its initial public offering (IPO) is reportedly targeted for November, though Anthropic hasn't confirmed a date. OpenAI is reportedly raising at least $30 billion at a valuation of about $1.4 trillion, with yearly revenue approaching $70 billion.

Two readings. Some market commentators argue that light regulation clears the way for investors. Other critics argue the reverse: that Anthropic's safety warnings are a tactic to create urgency and raise its value before the IPO.

No one has put a dollar figure on what the accord itself is worth to these companies. But a voluntary pledge with no fines keeps Washington's hand light at a moment when trillions of dollars in valuations depend on investor confidence.

Disclosure: Anthropic makes Claude, the AI tool used to help research this newsletter.

While Washington relies on promises, some states have written actual laws. California's Senate Bill 53 (SB 53), in effect since January 1, requires the biggest AI developers to publish safety plans, report serious incidents, and protect whistleblowers, with fines of up to $1 million per violation. New York City requires bias audits of AI hiring tools, and Illinois bars employers from using AI in ways that discriminate.

Enforcement is uneven: a state audit found New York City received just two complaints in its law's first two years. Washington is also pushing back, with a Justice Department task force set up to challenge state AI laws. But 36 state attorneys general from both parties oppose a federal override, and so far no court has ruled that federal authority overrides a state AI law. That's a second race to watch: states writing rules, and a White House working to limit them.

The case for vanguards. Leaders of most of America's major AI labs have now committed in writing, side by side, to outside audits and board-level oversight. Amodei and Zuckerberg both described the accord as a start, and Zuckerberg said it isn't meant to be the only thing the industry does. Johnson argued that Congress shouldn't overreact, because the AI labs are ahead of lawmakers on the technology. Supporters also argue that heavy rules could hand China the lead.

The case for skeptics. House Democratic leader Hakeem Jeffries called an unenforceable pledge "not the right response at this moment," given that industry leaders themselves are warning of serious harm. Senator Mark Warner, a Virginia Democrat, wants testing required by law before the most powerful models launch, and Republican Senator Thom Tillis doubted AI regulation could be settled in one meeting. University of Michigan professor Nazanin Andalibi argued that self-regulation lets the most powerful people in the room define both the problem and the solution.

The accord is built to prevent catastrophes: hacking, bioweapons, chemical threats. It doesn't directly address what happens to your personal information when AI systems can reach it, and the Medicare incident shows that risk isn't hypothetical.

Your data matters to the business side, too: OpenAI has already experimented with digital advertising, a business model that runs on personal data.

For now, the race continues at full speed, and the guardrails depend on companies keeping their word. That makes your choices matter: which AI tools you use, what you let them reach, and what you ask of the people you vote for.

Morally binding. A promise someone is expected to keep because it's the right thing to do, not because a law requires it. There's no penalty for breaking it.

Independent audit. An outside group checking whether a company is doing what it says. It only works if the auditor has real access and the results are made public.

Codify. To turn a voluntary practice into an actual law or regulation.

Preemption. When a federal law overrides state laws on the same subject. It's the heart of the fight between Washington and the states over AI.

Worth noticing: a pledge can be sincere and still not be enough. Both can be true at once.

If the companies building AI write the rules, run the checks, and grade themselves, what would it take for you to trust the result? And if the race is the reason not to slow down, who decides when it's been won?

Vanguards or villains? Reply and tell us.

POLL: Is a voluntary pledge from AI companies enough to keep AI safe?
→ Yes / It's a start / No / I'm not sure

Next time: When AI gets it wrong about you, do you have a right to an explanation and an appeal? We'll show you how to ask why.

Sourcing: ABC News (Australia), Axios, CBS News, CNBC, CNN, NBC News, NPR, Roll Call, TechCrunch, the Washington Times, Forbes, and the White House accord, plus state law records and trackers for California, New York City, and Illinois. Contested claims are noted. Disclosure: Anthropic, one of the accord's signatories, makes the AI tool used to help research this newsletter.

— Ezekiel Sage, Editor-in-Chief

THE CLOUDED TRUTH — no jargon, no hype, no panic