President Trump signed Executive Order 14409 on June 2, 2026110. It offers AI companies a trade. Hand over your most powerful model for up to 30 days, then ship it with federal cover12. A classified process, run by the National Security Agency, decides which models qualify2.
That process cleared its own deadline on paper. The rulebook it produced has stayed closed to the public for 138 days. A court filing now sets Oct. 30 as the date the government must show at least part of it89.
Three readers feel this first. A security lead must guess whether a product counts as covered. Her lawyer finds a rule the government keeps classified. A policy team budgets for a review with an unpublished scope.
What "voluntary" actually asks for
The order stays voluntary by design. A company may choose to share a "covered frontier model" with federal reviewers before release, then route it to a short list of "trusted partners"2. It explicitly rules out a license requirement for any new model110. Thirty days, then the company decides what comes next1.
Skipping the review carries a market cost, outside any legal one. The order binds through choice, so Congress and customers are the only enforcers on hand12.
One word does the gatekeeping: covered. A classified benchmark decides what crosses that line. The National Security Agency built it, working with the National Cyber Director, the president's science adviser and the Pentagon2. That deadline landed Aug. 1, 20262.
Treasury and the Cybersecurity and Infrastructure Security Agency got a parallel 30-day clock to stand up a vulnerability-sharing clearinghouse12. A draft of the order had asked for 90 days of pre-release access. David Sacks, the White House's AI adviser, pushed that number down to 30 before signing1.
A model went dark before the deadline even started
Ten days after the signing, the government moved fast, and outside any formal review. Anthropic's own account says the Commerce Department applied export controls to Claude Fable 5 and Mythos 5 on Friday, June 124. Officials had learned that Amazon researchers found a way around Fable 5's safeguards.
They simply asked it to list software vulnerabilities4. Anthropic could only guess at a user's nationality in real time, so it cut off both models worldwide rather than risk a violation4.
Commerce moved fast to unwind its own order. The government cleared Mythos 5 for a limited set of vetted American organizations on June 264. Commerce lifted the broader order late on June 30, and Fable 5 came back to every user the next day45. Anthropic thanked its users on X for the three-week wait, writing that it was "grateful to our users for their patience, and to everyone who worked with us on redeploying the models"5.
138 days separate the order's signing from any public rulebook
Executive Order 14409 and the events that followed, June 2 to Oct. 30, 2026
Each dot is one dated event; the magenta dot at the end is the court-ordered release date, still ahead when this piece runs.
Sources: CyberScoop [1]; IAPP [2]; Anthropic [4]; Al Jazeera [5]; Protect Democracy [9]. Chart by The AILately.com Desk.
The numbers behind this chart
| Date | Event |
|---|---|
| June 2, 2026 | Trump signs EO 14409 |
| June 12, 2026 | Export curbs on Fable 5. Commerce restricts Fable 5 and Mythos 5 |
| July 1, 2026 | Fable 5 returns. Fable 5 returns worldwide |
| Aug. 1, 2026 | 60-day benchmark due. NSA benchmark deadline; the White House says it was met |
| Sept. 1, 2026 | Protect Democracy sues. Sues for the framework's text |
| Oct. 30, 2026 | Oct. 30 release ordered. Court filing sets this release date |
That gap runs four months, from one public document to a court filing. It frames the fight over what the government owes the public once it holds this kind of leverage.
The case for a classified rulebook
Sacks made the administration's defense in public, posting on X the day of the signing. He wrote that the order is "completely consistent with the discussions that I have participated in," adding that it was meant to "apply only to models that represent a meaningful step-change in cyber capabilities" rather than every new version number1. Sacks called the shortened window "a game changer," arguing it let frontier labs comply while keeping new releases on schedule1.
Read plainly, Sacks is arguing that speed and secrecy serve the same goal. A published threshold, on his logic, hands a lab's rivals, and foreign governments, a map of exactly what to build around. Classified lines, in this telling, catch a true leap in capability and let routine updates ship on schedule.
The case against secrecy
Rep. Don Beyer, Democrat of Virginia, co-chairs the Congressional AI Caucus. He read the order the day it landed and called it thin cover. "The risks posed by frontier AI models and their advanced cyber capabilities are real and demand serious attention, but this executive order falls short of providing a credible framework to manage those risks," Beyer said3.
By September, the complaint had shifted from the order's design to its silence. Rep. Lori Trahan argued the whole approach belongs somewhere Congress can watch it, saying AI governance "belongs in a civilian agency, where it can be seen and questioned, not buried inside the national security apparatus"7. Tim Harper of the Center for Democracy and Technology put the gap in plainer terms: "The public has a right to know what the government's interpretation of the law is," he said6.
The record complicates a tidy compliance story too. Reporting names five labs at an Aug. 4 White House briefing on the framework: OpenAI, Anthropic, Google, Meta and Nvidia7. Only one has gone on the record with an actual submission. Sam Altman told reporters OpenAI ran its Astra model through the review and called the process productive6.
Only 1 of 5 labs at the briefing has a submission on the record
AI labs named at the Aug. 4 White House briefing on the framework, by whether a submission is confirmed, as of Oct. 18, 2026
Each cube is one lab the record ties to the briefing; the magenta cube is OpenAI, the only one that has confirmed on the record that it ran a model through the review.
Sources: CBS News [6]; Fortune [7]. Chart by The AILately.com Desk.
The numbers behind this chart
| Item | Value |
|---|---|
| Confirmed a submission: Astra | 1 |
| Named, pending confirmation | 4 |
| Whole | 5 |
The government told reporters in early August that it had finished the benchmark and the framework on schedule6. It declined to publish either document, name who had seen them, or cite the legal authority behind them6. Protect Democracy sued the Office of the National Cyber Director, Treasury, Commerce and the White House science office on Sept. 1 to force disclosure under the Freedom of Information Act9.
A joint filing in that case commits the government to releasing the framework's text by Oct. 30. That date reaches only the clearinghouse records named in the suit. Any side deal a lab struck later sits outside it89.
What to do
A security or legal lead needs a published standard to certify compliance against, and the government keeps that standard closed. Build a file instead: every model your company has shared with a federal reviewer, the date of each exchange, and the response it drew. That record is the only leverage left once Washington decides what counts as covered.
Watch the procurement contracts too. Adam Conner's reporting for Lawfare argues that federal buying power will end up setting the real bar, beyond the order's text8. A vendor chasing government business will face the benchmark regardless of a statute8. Legal teams should ask their sales side whether any contract already references the framework by name.
By the numbers
- 30 days is the longest a company holds off release for a covered model, down from a 90-day draft1.
- A classified threshold had 60 days to define a "covered frontier model," a deadline that landed Aug. 1, 20262.
- 138 days separate the order's June 2 signing from this piece's Oct. 18 publication, with the rulebook still closed to the public1.
- Claude Fable 5 sat under an export-control freeze for 19 days, from June 12 to July 145.
- Five AI labs attended an Aug. 4 White House briefing on the framework's rollout7.
- One lab, OpenAI, has confirmed on the record that it ran a model, Astra, through the review6.
- Oct. 30 is the date a court filing sets for releasing the framework's text, covering one slice of the program89.
What to watch
Oct. 30 will show whether the government keeps its word to a federal court. That bar differs from the deadline it already claims to have cleared. Congress holds a second lever.
Sen. Josh Hawley wants his bill with Sen. Richard Blumenthal to make the reporting mandatory, saying flatly, "I would go farther. I think we ought to enact my legislation that I have with Senator Blumenthal that would make that sort of reporting and monitoring mandatory"11. A floor vote on that bill would settle whether this stays a handshake or becomes law.
Watch, too, whether any lab beyond OpenAI puts a submission on the record. A second name would test whether the process works for more than one company already talking about it in public.
Sources
- Derek B. Johnson, "Trump administration releases scaled-back AI executive order," CyberScoop, June 2, 2026, https://cyberscoop.com/donald-trump-white-house-ai-executive-order-scaled-back/
- Alex LaCasse, "Unpacking the White House executive order on frontier AI, cybersecurity," IAPP, June 3, 2026, https://iapp.org/news/a/unpacking-the-white-house-executive-order-on-frontier-ai-cybersecurity
- Office of Rep. Don Beyer, "AI Caucus Co-Chair Beyer Response to Trump's Latest AI Executive Order," Office of Rep. Don Beyer, June 2, 2026, https://beyer.house.gov/news/documentsingle.aspx?DocumentID=9132
- Anthropic, "Redeploying Claude Fable 5," Anthropic, June 30, 2026, https://www.anthropic.com/news/redeploying-fable-5
- John Power, "US lifts restrictions on Anthropic's powerful AI models Fable and Mythos," Al Jazeera, July 1, 2026, https://www.aljazeera.com/economy/2026/7/1/us-lifts-restrictions-on-powerful-ai-models-fable-mythos-anthropic-says
- Kathryn Watson and Olivia Rinaldi, "White House framework for testing AI models remains hidden as concerns about threats mount," CBS News, Sept. 10, 2026, https://www.cbsnews.com/news/ai-model-framework-white-house/
- Beatrice Nolan, "Washington is keeping its AI rulebook private. Smaller AI labs aren't happy.," Fortune, Aug. 6, 2026, https://fortune.com/2026/08/06/washington-is-keeping-its-ai-rulebook-private-smaller-ai-labs-arent-happy/
- Adam Conner, "The Many Open Questions About the Trump Administration's New De Facto AI Model Licensing System," Lawfare, Oct. 13, 2026, https://www.lawfaremedia.org/article/the-many-open-questions-about-the-trump-administration-s-new-de-facto-ai-model-licensing-system
- Protect Democracy, "Uncovering the Trump administration's secret rules for AI model release," Protect Democracy, Sept. 15, 2026, https://protectdemocracy.org/work/uncovering-the-trump-administrations-secret-rules-for-ai-model-release/
- The White House, "Fact Sheet: President Donald J. Trump Promotes Advanced Artificial Intelligence Innovation and Security," The White House, June 2, 2026, https://www.whitehouse.gov/fact-sheets/2026/06/fact-sheet-president-donald-j-trump-promotes-advanced-artificial-intelligence-innovation-and-security/
- Allison Mollenkamp, "Executive order sets voluntary cyber reviews for advanced AI," Roll Call, June 2, 2026, https://rollcall.com/2026/06/02/executive-order-sets-voluntary-cyber-reviews-for-advanced-ai/