Photo by Andrew MacDonald on Unsplash
Thirty-four months.
That is the distance between the Biden administration's October 2023 executive order on artificial intelligence — the one that required safety testing for large AI models — and today, August 4, 2026. In that span, U.S. federal AI policy has been written, contested, unwound, and reopened, with Silicon Valley recalculating its compliance posture each time. According to Google News, which distributed the New York Times report underlying this discussion, the White House has once again reversed direction on AI rules, in a way that unsettled not only the companies being regulated but the administration's own earlier stance.
The 34-Month Whipsaw
A caveat belongs up front, because it changes how you should read everything below. As of August 4, 2026, direct retrieval of the current reporting failed: the New York Times and Reuters pages were access-blocked, and Bloomberg's technology section returned a 403 error. No current statistics, no fresh analyst commentary, and no primary government data could be pulled. What follows is therefore structural analysis built on facts that are firmly on the record — the October 2023 executive order and its safety-testing requirement for large models, the Trump administration's documented preference for technology-sector deregulation, and the existence of competing federal AI policy proposals aimed at the same companies — not a paragraph-by-paragraph retelling of a story this analysis could not open. Anyone who needs the operative language should go to the Federal Register text itself rather than to any secondary account, including this one.
What can be examined without the day's specifics is the shape of the pattern. And the shape is the interesting part.
The Mechanism: Volatility Costs More Than Strictness
The surface framing of this story is regulation versus deregulation — a familiar left-right axis that most coverage settles into within two paragraphs. That framing misses the variable that actually governs corporate behavior. For a company deciding whether to spend on model evaluations, red-team staffing, audit logging, and outside counsel, the binary that matters is not strict versus loose. It is stable versus unstable.
Consider the arithmetic of a compliance build. A firm that stood up a model safety-testing function in response to the October 2023 order was making a multi-year capital commitment against an expected rule set. If that rule set holds, the spend amortizes; the capability becomes an operating cost and, eventually, a competitive asset. If the rule set is withdrawn, the same spend is a write-off. If it is withdrawn and then partially reinstated in different form, the firm pays twice — once to build, once to rebuild — and the second build is more expensive because institutional confidence in any future mandate has fallen. The second-order effect is that rational firms start under-investing in compliance capability on purpose, holding the option to wait. Under-investment then makes any eventual enforcement more disruptive, not less.
Break the 34 months into their two segments and the planning problem becomes visible. Roughly 15 months separated the October 2023 order from the late-January 2025 reporting on the incoming administration's AI rules, and roughly 19 months separate that moment from today. Neither interval is long enough to complete, validate, and depreciate an enterprise-scale governance program.
Chart: Elapsed months between dated U.S. AI policy milestones, computed from the October 2023 executive order and reporting dated January 2025, measured to August 4, 2026.
A careful skeptic will push back here, and the pushback deserves a straight answer. The objection: deregulation is simply good for AI equities, oscillation is noise, and markets price the destination rather than the route. That is half right. Markets do price expected end states — but only when an end state is knowable. When the governing rule can invert with each administration, there is no terminal value to discount toward; there is a probability distribution, and distributions widen the further out you look. The practical consequence shows up not in stock market today headlines but in enterprise procurement: buyers of AI systems in regulated industries delay signing when they cannot tell which certification will be demanded of them in eighteen months. Delayed signatures are delayed revenue, and delayed revenue is a real number even when the policy debate produces none.
Photo by Maxence Pira on Unsplash
Who Wins Under Which Condition
Three scenarios, three different winner sets — and the third is the one most coverage skips.
Under durable deregulation, the advantage flows to whoever can spend fastest on capability. Frontier labs running capital-intensive training cycles get to convert compute directly into product without a testing gate in the middle. The losers are the compliance-tooling vendors and audit firms whose addressable market was created by the October 2023 mandate.
Under durable re-regulation, the advantage inverts — but not toward the obvious party. It flows to the large incumbents that already built safety-testing infrastructure for the 2023 order, because their sunk cost becomes a barrier for everyone else. This is the counterintuitive part: strict rules, held steady, are a moat for the biggest firms. Startups lose, not because the rules are unfair, but because fixed compliance costs do not scale down.
Under sustained oscillation — the condition actually observed across these 34 months — neither group captures the advantage. It accrues instead to businesses whose economics are indifferent to the rule set: semiconductor and infrastructure suppliers selling into training demand regardless of who audits the output, cloud providers billing by the hour, and the legal and policy-advisory layer that gets paid on every reversal. Policy churn is a revenue event for the people who interpret policy. The moat compresses when compliance is cheap; it widens when compliance is expensive and predictable; and it simply relocates — toward capital and toward counsel — when compliance is expensive and unpredictable.
Readers tracking how political volatility transmits into asset prices will recognize the structure; it echoes the verification problem Smart Finance mapped in its analysis of pressure on the Federal Reserve, where the durable market variable turned out to be institutional predictability rather than the direction of any single decision.
Bottom Line
Our read, on balance: the most likely six-to-eighteen-month trajectory is not a decisive settlement in either direction but continued partial reversal, because competing federal AI policy proposals remain live and no single framework has yet survived a full corporate planning cycle. If that holds, the returns accrue disproportionately to the layers of the stack that get paid regardless of the rule set — and any investment portfolio built on a confident bet about which way Washington lands on AI is being paid for taking political risk, not technology risk. Those are different exposures, and conflating them is the most common error in this trade.
- The measurable fact: 34 months have elapsed since the October 2023 executive order's safety-testing requirement for large AI models, split into roughly 15- and 19-month segments by intervening policy shifts.
- The mechanism: instability, not severity, is what makes compliance spending unrecoverable — firms pay twice or defer entirely.
- The overlooked winner: steady strict rules would advantage large incumbents; churn advantages infrastructure suppliers and the advisory layer instead.
- The verification limit: as of August 4, 2026, the New York Times, Reuters, and Bloomberg pages for this story could not be retrieved (Bloomberg returned a 403), so treat all same-day specifics elsewhere as unconfirmed until checked against primary text.
- For readers: for personal finance and financial planning purposes, treat AI-policy headlines as a volatility input, not a directional signal.
Frequently Asked Questions
What did the October 2023 AI executive order actually require?
The order issued under the Biden administration required safety testing for large AI models. That is the specific provision on the record; the broader framework's current operative status as of August 4, 2026 should be confirmed against the Federal Register rather than any secondary summary, including this one.
Does AI deregulation automatically help AI stocks?
Not automatically. Looser rules lower near-term compliance costs, but they also erode the barrier that protects firms which already built testing infrastructure. Whether a given company benefits depends on whether its advantage comes from capability or from compliance scale — and on whether the change is durable. This is analysis, not investment advice.
How should long-term investors treat US AI regulation headlines?
Industry analysts generally distinguish between policy direction and policy stability. Direction moves individual names; stability moves the discount rate applied to the whole category. For financial planning horizons measured in years rather than news cycles, the second variable has historically been the more consequential one.
Disclaimer: This article is editorial commentary based on publicly reported facts and does not constitute financial, investment, or legal advice. No independent product testing was performed. Several primary sources were inaccessible at the time of writing, as noted in the body. Research based on publicly available sources current as of August 4, 2026.