USMCA and AI: How Mexico’s Governance Vacuum Risks Trade Gains

Pull up the US Commerce Department's list of 18 countries cleared for preferential access to advanced AI chips and export licenses, and you will find Canada sitting comfortably on it. Mexico is not there.

No public document spells out exactly why. But the analysts who track this closely point to the same explanation: Mexico's technology regulatory framework has not caught up to what Washington now treats as a prerequisite for trust. That is not a footnote. It is the clearest signal yet that artificial intelligence has stopped being a domestic policy conversation in North America and become a trade one, and Mexico is showing up to that negotiation without half its paperwork done.

The timing could not be worse, or better, depending on how fast the country moves in the next few weeks.

On July 1, the United States, Mexico and Canada convened the first mandatory joint review of the USMCA, a checkpoint written into the agreement six years ago almost as a formality. Nobody treated it like one this year. US Trade Representative Jamieson Greer told the other two governments that Washington would not renew the agreement in its current form. Mexico and Canada both said they wanted a full 16-year extension. Nobody got what they wanted, and the deal now shifts into an annual review cycle that will run until the three countries either agree to extend it or it expires on schedule in 2036.

None of that means USMCA is at risk of collapsing tomorrow. It stays fully in force while the parties keep talking, and a clean extension remains available at any point if Washington changes its mind. But "fully in force" is doing a lot of work in that sentence for anyone running a business that depends on predictable rules. Bilateral talks between the United States and Mexico have already run three rounds this year, in May, June, and July, working through automotive rules of origin, steel and aluminum, and what negotiators keep calling economic security, which in practice means keeping Chinese capital and Chinese technology out of the regional supply chain. Canada has been mostly sidelined from that particular conversation, and Washington added a fresh 50% tariff on Canadian vehicles, alcohol, and dairy in August. A fourth US-Mexico round is already scheduled for Washington in September.

The Divide on Digital Trade

Buried inside all of that trade friction is a chapter almost nobody outside a handful of trade lawyers reads closely: Chapter 19, digital trade. It is the part of USMCA that lets a Mexican company store customer data on servers north of the border without building a local data center just to satisfy a foreign law, that stops any of the three governments from forcing companies to hand over source code, and that treats software crossing a border the same way it treats a container of auto parts. When it was negotiated in 2018, it was considered the most advanced digital trade framework any government had signed. Six years later, it was not built with artificial intelligence in mind, and everyone close to the review knows it.

Researchers at the Center for Strategic and International Studies, who have spent the past year briefing all three governments ahead of the review, argue the chapter's core provisions have aged well and should not be reopened wholesale. Their preferred fix is narrower: attach a separate AI protocol to the agreement, a side letter that would set shared definitions of AI risk, mutual recognition of safety standards between the three countries, and a standing working group to keep the rules current as the technology moves. The Inter-American Dialogue has run its own task force on the same question since last year and landed in a similar place. A policy group called ITIF has gone further still, proposing Chapter 19 be renamed entirely, from a digital trade chapter to a digital economy and security chapter, built explicitly to cover AI and to align export controls against China.

Mexico is not without its own ideas here. IMCO, the country's most respected competitiveness think tank, has argued Mexico should use the review to go on offense: build what it calls a regional economic security agenda, one that ties technology cooperation, modern border infrastructure, and supply chain resilience into a single strategic pitch rather than reacting country by country to whatever Washington proposes next. That is the right instinct. It is just difficult to make that pitch with any credibility when your own AI rulebook is still a stack of drafts sitting in a Senate committee.

Every one of those proposals, American, Mexican, or otherwise, assumes Mexico shows up to the table with something to negotiate. Right now, it does not.

Mexico does not have a general AI law. It came close this year. A Senate commission spent the spring saying it was preparing a comprehensive framework, and in April a separate bill landed in the Chamber of Deputies proposing to amend Article 73 of the constitution, the provision that would actually give Congress the authority to pass a national AI statute in the first place. Neither made it to a floor vote before the legislative session closed. Sen. Karina Ruiz introduced her own version in February, which would create a national AI agency with real enforcement teeth: fines tied to the harm caused and the benefit gained by whoever broke the rules. That bill is still sitting in committee too.

Patchwork of Reforms, Rules

What exists instead is a patchwork. Reforms to federal labor law and copyright law touching AI use took effect in May. Data protection rules now require companies to disclose when they are using automated decision systems and give people the right to ask for human review. None of it adds up to the kind of framework that lets Mexico walk into a trade negotiation and claim a credible seat at the table on AI governance. And the rest of the world is not waiting for Mexico to catch up. The European Union's AI Act enters its mandatory deepfake labeling phase this August, which means any Mexican exporter with European customers now has to comply with European AI rules while their own government still has not decided what its rules will even be.

To be fair, there is a real argument for moving carefully. Rushing a single sweeping law modeled on the European Union's, with its rigid risk tiers and heavy compliance costs, could easily backfire on a country where small and midsized companies still do most of the hiring. Industry voices such as Mexico's tech trade association AMITI have pushed back gently against copying Brussels, arguing a sector-by-sector, risk-based approach can work better for an economy at this stage. Fair enough. But caution and paralysis are not the same thing, and paralysis now has a deadline it did not have six months ago.

The Gap in the Numbers

The gap shows up in the numbers too. A study built on INEGI's 2024 economic census found that only 4.8% of Mexican manufacturing companies with more than 10 employees use AI in any form. Across every sector, the average sits at 8%. The OECD average is 19.1%. A separate AWS-commissioned study released this year found something closer to half of Mexican companies, 48%, now use AI in some capacity, up sharply from 38% the year before. The gap between those two figures tells its own story: adoption is climbing fast in services and finance, while manufacturing, the sector nearshoring was supposed to supercharge, is barely moving. And of the companies that have adopted AI at all, nearly two-thirds are still stuck running chatbots and basic document automation, which is nowhere close to the agentic systems already handling entire workflows elsewhere.

None of this is abstract if you run a company that depends on any of it. If you sell into the United States or Canada, the rules governing cross-border data are being renegotiated right now, and Mexico has no domestic framework backing its position at the table. If you are trying to buy or license advanced AI hardware, you are already operating in a country the US government has not cleared for preferential access, for reasons tied directly to that same regulatory gap. If you serve even one client in Europe, you are already complying with foreign AI rules your own government has not matched. Mexico's foreign direct investment confidence climbed six spots this year, to 19th globally, largely on the strength of nearshoring. Kearney and Morgan Stanley, the two firms that track this most closely, both attached the same caveat to that good news: the ranking assumes the USMCA review produces clarity, not chaos. An AI governance vacuum sitting inside a bigger trade fight is exactly the kind of thing that turns clarity into chaos.

The uncomfortable truth is that Mexico still has time, but not much of it. The fourth round of talks in September will not wait for the Senate to finish arguing over Article 73. Canada's own AI transparency consultation closes that same month. Every week Mexico spends without a functioning AI framework is a week the other two governments, and every serious economy negotiating with them, get to write the rules of the road without a Mexican signature on the page. The companies that will feel this first are not the multinationals with legal teams in three countries. They are the Mexican manufacturers, exporters, and founders who assumed the rules would eventually catch up to them. This time, the rules are not going to wait. Mexico has to catch up to the rules, and it has weeks, not years, to decide whether it wants a hand in writing them or a copy of whatever everyone else agreed to without it.