Not a Pause, but Guardrails: Shapiro Plants His Flag on AI Policy in Pittsburgh

One week after the House adjourned early rather than confront AI-doom warnings, one of the Democratic Party’s most-watched governors walked into a Pittsburgh conference hall and tried to answer the question Congress left hanging: what should the federal government actually do about artificial intelligence?

Pennsylvania Gov. Josh Shapiro used a Thursday evening keynote at the AI Horizons summit in the city’s Bakery Square neighborhood to call for “strict guardrails” — swift federal regulation built around independent third-party oversight of AI companies, mandatory transparency, and a genuine prioritization of safety. But he drew a hard line against the pause camp: regulate as you build, not instead of building.

“I am not a doomer,” Shapiro told the audience, preempting the easiest label in politics. “I don’t think it is inevitable that AI is going to take over the world or that someone will misuse it to kill us or to do harm. But I do have real concerns, as do many Pennsylvanians and many Americans.”

Why a governor is talking about federal AI rules

The speech lands at a peculiar moment in the AI debate. Researchers and leaders at Anthropic, OpenAI and Google have spent the past two weeks sounding public alarms about the risks of their own frontier systems — a wave of warnings that culminated in Congress getting closed-door briefings from Hinton, Tegmark and Cotra, and in Speaker Mike Johnson sending the House home early rather than schedule votes on guardrail legislation.

Into that vacuum walks Shapiro, a swing-state governor with visible 2028 ambitions, whose track record on AI is unusually concrete for a politician:

• August 2026: He signed an executive order tightening rules on data-center development in Pennsylvania, after local opposition to AI infrastructure projects turned fierce.
• May 2026: His administration sued a generative AI startup whose chatbots posed as licensed medical professionals and dispensed unlawful medical advice.
• February 2026: He created an AI Task Force to crack down on AI companion bots that misrepresent themselves as licensed professionals.

As Axios noted, the arc is striking: a governor who spent 2025 and early 2026 promoting AI investment in his state has pivoted to regulating it, tracking a broader shift in public opinion as data centers became a liability in local politics and lab insiders began warning that their own systems could cause catastrophic harm.

“Flat-out wrong” to rely on tech companies

The sharpest line of the night was aimed not at the AI industry but at the hands-off governing philosophy that has kept federal rules stalled. Relying on AI companies to police themselves, Shapiro argued, is “flat-out wrong” — and the events of the past several weeks, from undisclosed agent incidents to open letters from lab staff, have made the warning signs impossible to ignore.

“We cannot ignore the warning signs, nor downplay the risks,” he said, while insisting the underlying opportunity for Pennsylvania — a state angling to be an AI hub between the East Coast corridor and the Midwest — remains real.

He also swiped at both parties’ favored deflections. Blaming China for souring American sentiment toward AI, as some Republicans like Sen. Dave McCormick have done, is “reductive” in his view. And the Trump administration’s instinct to isolate the United States from international cooperation, he argued, is actively hurting the prospects of ensuring AI develops responsibly worldwide — a notable point given that AI safety, unlike most tech policy, has a meaningful diplomatic track running through US-China channels.

The false choice

The core of Shapiro’s argument is that the current debate offers a false binary. On one side is the position he attributes to President Donald Trump: ignore the risks and go full steam ahead, dismissing safety talk as a hoax. On the other is the growing pause movement, energized by researchers who quit frontier labs warning that companies are “gambling with our lives.”

Shapiro rejected both. It is possible, he told the crowd, to regulate AI and still have it succeed — to build the plane and the aviation authority at the same time. That framing puts him closer to the “guardrails without brakes” position that lab leaders like Anthropic’s Dario Amodei have advocated than to the hard-pause camp, while still going further than most elected Democrats have been willing to go on federal mandates.

The political subtext is hard to miss. NBC News framed the speech as a marker among potential 2028 contenders, who are now staking out positions on an issue that barely existed in the last presidential campaign. With the midterms approaching and polling showing public anxiety about AI rising — Pew’s latest survey found Democrats now outpace Republicans as the party more worried about AI, 56% to 49% — being early with a coherent AI position may matter the way being early on the internet economy mattered in 1999.

What he actually wants Congress to do

Stripped of the 2028 positioning, the ask is fairly specific:

1. Independent third-party oversight of AI companies, analogous to financial auditors or aviation inspectors, rather than the current self-policing model.
2. Increased transparency from labs about capabilities, incidents, and risks — something the labs themselves have begun volunteering in limited form (OpenAI’s new misalignment disclosure framework, Anthropic’s R&D automation metrics) but which Shapiro wants mandated.
3. Safety prioritization baked into federal policy, including how compute and infrastructure incentives are structured.

Notably absent from his list: a pause, licensing regimes, or new agencies. The agenda is deliberately incremental — oversight, transparency, safety — which may be the only kind of AI bill that can plausibly pass in the current Congress, if any can.

Analysis: the swing-state calculus

There is a cynical read and a charitable one. The cynical read: a governor whose state was burned by data-center politics saw which way the wind was blowing and repositioned accordingly, three years before a presidential primary. The charitable read: Pennsylvania’s experience — an AI infrastructure boom colliding with local backlash, a fraudulent medical chatbot lawsuit, and constituent anxiety — is exactly the empirical education in AI governance that members of Congress, insulated by fundraising cycles and four-day workweeks, have not gotten.

Both can be true. What matters for the policy debate is that Shapiro’s Pittsburgh speech is among the first complete statements from a top-tier national Democrat of what “AI regulation” concretely means: no pause, no self-regulation, and no pretending the choice is between doom and dominance. With the House having just demonstrated it would rather adjourn than legislate, and the Senate holding closed briefings on extinction risk, a governor filling the vacuum with specifics is — for now — the most substantive game in town.

The next test is whether anyone in Washington picks it up.