White House order aims to promote artificial intelligence innovation and security
Most technology rules arrive late. They get written after the damage is already done, named for the disaster they were meant to prevent.
Governments are not built to move first. They move when something scares them.
For roughly two years, the United States treated artificial intelligence (AI) exactly that way. The official posture was hands off. Let the labs build, let the models ship, and write the rules later, ideally much later.
The bigger worry in Washington was never really that AI might be dangerous. It was that slowing down American companies would hand the lead to China. So nobody wanted to be the one to tap the brakes.
That held right up until one model did something unsettling enough to change minds at the top. Then, the White House stopped waiting. President Donald Trump signed an executive order asking the most powerful AI companies to show the government their newest models before the rest of us are allowed to touch them.
The order is voluntary, for now. The machinery behind it is not small, and the reason it exists is the part worth your attention.
What the White House's new AI order actually asks for
When I reviewed the order, which is titled "Promoting Advanced Artificial Intelligence Innovation and Security," the standout feature is a swap. AI labs can give the government early access to their most capable models, and federal agencies get to study them before the public does.
Labs would share these "covered" frontier models for up to 30 days before a public launch, according to The Hill. Taking part is voluntary, and no model has to pass a review before it ships.
More AI:
- Micron sits at the center of a red-hot chip rally
- IBM CEO sends blunt message on AI and quantum computing
- Anthropic CEO makes shocking admission about AI
The other half of the order is pure cybersecurity. It directs federal agencies to harden the systems behind national security, the Department of War, and civilian government, and to expand cyber hiring, according to the White House.
There is real institutional muscle under that "voluntary" label. The framework leans on classified benchmarks run by the National Security Agency (NSA) and a government-managed window to review models before release, the law firm Ropes & Gray noted.
My read is that the government is quietly building a permanent on-ramp to the most powerful AI, then calling it optional so nobody panics. For you, it means the next big model your employer rolls out, or the one running inside the app on your phone, may clear a federal checkpoint first.
Why Mythos Preview rattled the White House
None of this happens without a scare, and the scare has a name. It is Mythos Preview, a frontier model from the AI company Anthropic.
In April, Mythos Preview demonstrated it could autonomously hunt down thousands of severe software flaws on its own, including in widely used operating systems and browsers, according to NBC News.
The scale is what spooked people.
- Mythos Preview flagged more than 23,000 software vulnerabilities, including over 6,000 rated high or critical severity, with about 91% later judged valid by independent firms, Security Boulevard indicated.
- Anthropic first gave roughly 50 partners early access through a program it calls Project Glasswing, then expanded that to 150 organizations across more than 15 countries on June 2, according to NBC News.
- The new order asks AI labs to share covered models with the government for up to 30 days before a public release, The Hill reported.
The fear was not subtle. A tool that good at finding holes in software could just as easily be pointed the wrong way.
A "bad actor could use Mythos to target various cybersecurity vulnerabilities," Vice President JD Vance said at a news conference in May, according to NBC News.
The president nearly signed a version of the order that month, then pulled it at the last minute over fears it would slow American labs while China kept racing, NBC News noted. The June 2 order is what survived that fight.
What Sam Altman wants out of the deal
The morning after the signing, the industry showed up in person. OpenAI CEO Sam Altman spent June 3 meeting with White House officials and lawmakers from both parties, CNBC confirmed.
His public verdict was warm. "The new EO gets the balance right," Altman wrote on X (the former Twitter), backing an approach that keeps building the best models while putting cyber tools in the hands of defenders.
Related: The pros and cons of using AI for your taxes
The private agenda was sharper. Altman pushed for more transparent disclosure of how the government tests models, and for a larger role for OpenAI in shaping the process, according to Bloomberg.
He worked every corner of the Capitol, sitting down with House Speaker Mike Johnson and House Minority Leader Hakeem Jeffries, plus Senator Bernie Sanders, CNBC reported. When Sanders and Johnson take the same meeting, the company in the middle is not worried about its access.
Where the real AI security risk still lives
Here is the part the order does not solve, and the part I keep coming back to. Testing a model before launch tells you how it behaves in a lab. It says almost nothing about what that model does once it is wired into a real company.
"Pre-release vetting is necessary, but it tests a model in isolation," said Matan Bar-Efrat, chief executive and co-founder of Rein Security, which secures AI agents running inside large enterprises. "It can't test what the model does once it's wired into your APIs, your data stores, and running live in production."
His point lands for anyone whose job now runs on AI. The risk most people will actually meet is not a rogue frontier model. It is an ordinary, approved model plugged into company systems with too much access and not enough oversight.
That is the gap between what Washington is testing and where breaches happen. The order inspects the engine. The crash tends to happen out on the road, after the car is sold.
What the U.S. government AI order means for your money and your job
For investors, the order is less a brake on the AI trade than a new variable. A federal review window, even a voluntary one, is one more thing that can delay a launch or reshape a product.
For everyone else, the takeaway is simpler. The AI showing up in your work software, your bank, and the apps on your phone now passes through a government checkpoint that did not exist a month ago, built because one model proved how much damage a capable enough machine could do.
Right now the labs are cooperating, because cooperating is good politics and good marketing.
The real test arrives with the next Mythos, the model too capable to wave through. Whether "voluntary" survives that moment is the thing worth watching.
Related: Analyst resets Workday stock forecast on hidden AI risk
The Arena Media Brands, LLC THESTREET is a registered trademark of TheStreet, Inc.
This story was originally published June 5, 2026 at 6:03 PM.