Labor Secretary Says AI Has Not Displaced Workers, But Local Data is Unavailable

Puerto Rico’s secretary of Labor and Human Resources told Caribbean Business in an exclusive interview that the AI boom reshaping workplaces worldwide has not led to a notable displacement of local workers, and that the island’s employers and universities have already adapted to the technology faster than the public debate suggests.

“We’ve discussed these issues with various organizations and haven’t seen any displacement of workers by AI,” Secretary María del Pilar Vélez Casanova asserted. “On the contrary, there are jobs in Puerto Rico, and people are being hired every day.”

It is an unusually confident assessment, and it arrives at a moment when the same claim is being tested, and in places contradicted, by data elsewhere in the United States.

In Vélez Casanova’s telling, AI has generated a training obligation rather than a displacement problem. Employers, she said, have begun running programs to prepare existing staff for technological change, particularly older workers, and in some cases to prepare people they have not yet hired.

Those initiatives extend to workers reentering the labor force, she said. The Labor Department has worked with the Department of Economic Development and Commerce and with programs tied to the federal Workforce Innovation and Opportunity Act to expand training, including for older adults.

She was firmer still on education. Since taking office, she said, one of her recurring conversations with universities has been about adapting programs to labor-market demand, and the institutions have told her the work is already underway.

“In those conversations, they’ve told me they’re already integrating it, not only at the postsecondary level, but we’ve seen that high school teachers are incorporating artificial intelligence and helping their students use [AI tools] and learn the right way to work with AI. I believe AI has been gradually integrated into the curriculum.”

The secretary did flag one risk directly: employers using AI in recruitment need human oversight to stay inside the law.

“It is very important that human resources comply with the law, do not discriminate, that a proper evaluation is being conducted,” she said. “There must be staff in place to verify that the platform is indeed operating without violating any laws or current labor regulations in Puerto Rico.”

What the national data shows

The secretary’s core claim, that AI is not producing broad displacement, is consistent with the best available U.S. research. What that research also finds is that the effect is not broad. It is concentrated, and it is concentrated precisely among the youngest workers.

Economists at the Stanford Digital Economy Lab, working with anonymized ADP payroll records covering millions of U.S. workers, reported last month that employment among workers aged 22 to 25 in the occupations most exposed to generative AI now sits roughly 19% below where it would be had it tracked their peers in less-exposed work. That gap has widened from about 13% when the researchers first documented it in 2025.

The pattern runs almost entirely through reduced hiring rather than layoffs, which is why it can persist without ever appearing as a displacement event. Workers already in those jobs largely keep them. The jobs simply stop being offered to new graduates. ADP has recorded 34 consecutive months of year-over-year employment contraction for that age group in highly exposed occupations.

The Stanford authors, Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, are explicit that their findings are descriptive rather than causal, and they find no evidence of economy-wide AI job destruction. On that broader point, the secretary’s optimism has support. On entry-level hiring, the evidence points the other way.

How other jurisdictions are approaching it

The more consequential question raised by the secretary’s statement may not be whether displacement has occurred, but whether Puerto Rico currently has any instrument that would detect it if it had.

In May, California Governor Gavin Newsom signed Executive Order N-6-26 described by his office as first-in-the-nation, directing state agencies to prepare for AI-driven labor disruption. It requires the Employment Development Department to launch a public dashboard within 90 days showing AI’s employment impact by sector, drawing on unemployment insurance data.

The Order gives the state’s Labor and Workforce Development Agency 180 days to recommend revisions to California’s WARN Act so that layoff notices function as an early-warning system for AI-driven workforce changes and to review safety-net policies for displaced workers.

Connecticut and New York have already amended their own WARN statutes to require employers to tell the state when layoffs are attributable to the adoption of AI.

Puerto Rico has no comparable reporting requirement, no sector dashboard, and no published methodology for identifying AI-attributable job losses. Vélez Casanova’s conclusion that no displacement has occurred rests, by the secretary’s own description, on conversations with employer associations rather than on a monitoring system designed to surface it.

The graduate-readiness gap

The secretary’s most striking claim concerns schools, and it is the one most at odds with national survey data, though no equivalent Puerto Rico-specific study appears to exist.

A Handshake survey of 1,248 graduating seniors across nearly 500 U.S. institutions in March found that while 85% of the Class of 2026 used AI tools during college, only 28% received any formal instruction.

A separate Monster survey of more than 1,000 recent and impending graduates found just 36% believed their college had adequately prepared them to use AI in the workplace, while 40% said it had not.

The perception gap runs in a specific direction. Cengage Group’s employability research found that nearly nine in ten educators believe their students are workforce-ready, while roughly half of graduates say they feel unprepared even to apply for entry-level jobs in their field.

The secretary’s account of university readiness is, in effect, a report of what educators told her, which is the side of that gap that consistently runs optimistic.

Employers, meanwhile, are moving the requirement. Entry-level postings calling for AI skills nearly doubled year over year, according to Handshake data reported by CNBC, and beginning with this fall’s freshman class, Purdue University will require an “AI working competency” for graduation.

Reliance is the other side of integration

Whatever the displacement picture, one thing is not in dispute. Puerto Rico’s businesses have adopted these tools quickly. A July white paper from IslaIntel, a San Juan AI market-intelligence firm, found that 84% of organizations on the island have deployed AI in at least one business function, against a global baseline of 72%, though the firm noted adoption is uneven, with multinationals leading and many small and midsize firms still using only basic tools.

That reliance became visible on Thursday, when several of the platforms that adoption runs on went down at once.

OpenAI’s ChatGPT, Anthropic’s Claude and xAI’s Grok all suffered disruptions in the same 90-minute window Thursday morning, according to outage-tracking site DownDetector, with reports peaking above 37,000 for ChatGPT and near 1,300 each for Claude and Grok.

OpenAI told Newsweek the incident began with a routing error around 7:43 a.m. Pacific time and reported elevated errors across ChatGPT and its coding assistant, Codex. Anthropic’s status page initially listed several models before narrowing the problem to two.

Google’s Gemini, which runs on Google Cloud rather than Microsoft Azure, drew a spike in user reports but never confirmed an outage. Several outlets noted that the three affected platforms share Azure infrastructure, though no company confirmed a common cause. Services began recovering by late morning Pacific time and were fully restored by early afternoon.

For an island where 84% of organizations now run at least one business function on these tools, a two-hour interruption is a reasonable proxy for how deeply the technology has already been absorbed. The same integration that makes an outage disruptive is what makes the displacement question worth monitoring rather than settling by consultation.