The White House promised a tougher leash on powerful AI, but the leash is mostly made of paper, and paper has a habit of tearing when politics change.
What You Should Know
President Biden signed an AI executive order on October 30th, 2023, directing agencies to set safety standards and require certain model developers to share test results with the government. Executive orders can be rewritten or revoked by a future administration.
The order, formally titled “Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence,” was pitched as a way to move faster than Congress. It also created a convenient pressure point for the next president, who can claim to be pro-innovation while quietly deleting the guardrails.
The Executive Order That Can Disappear
In the document, the Biden administration leaned on existing authority, including the Defense Production Act, to push some AI makers to report safety test results and other information to the Department of Commerce when models meet certain risk thresholds.
It also tasked the National Institute of Standards and Technology, known as NIST, with shaping the technical backbone of what “safe” even means, and it told agencies across government to start knitting AI rules into procurement, cybersecurity, and civil rights enforcement.
That is the power move and the vulnerability. An executive order can marshal the federal bureaucracy quickly, but it can also be reversed quickly, especially when the policy is controversial, expensive to comply with, or easy to rebrand as red tape.
The Real Fight Is in the Details
Publicly, the debate sounds like values. Safety, jobs, security, and innovation. In practice, it is a knife fight over who bears the cost of testing, reporting, audits, and compliance, and who gets to call their system “safe enough” to ship.
For developers, mandatory reporting and standardized evaluations can mean more paperwork, more legal exposure, and more chances for regulators to say no. For the government, the reporting and standards work are the closest thing to visibility into models that can be deployed at scale before anyone outside the company understands their risks.
Meanwhile, Congress has held high-profile hearings on AI, but broad, durable legislation has been slow in coming, leaving executive action to do the heavy lifting. That vacuum is why a single election can matter as much as a single breakthrough model.
What to watch next is not just who wins the messaging war, but whether agencies actually lock rules into procurement contracts, regulatory guidance, and enforcement patterns that are harder to unwind quietly. The politics can flip overnight, but the paperwork trail can last longer.
