New York Governor Kathy Hochul has been talking about artificial intelligence in a way that’s both familiar and revealing: familiar because it’s framed as a tool for efficiency, and revealing because she’s describing it as something her administration is using not just for isolated tasks, but for a sweeping, almost bureaucratic-scale project—an attempt to scan and evaluate “every single rule, regulation, [and] policy” in the state.
The comments came during an interview with Bloomberg’s Odd Lots podcast, where Hochul said her team is using AI to help clean up New York’s legal code by identifying outdated provisions. In other words, the technology isn’t being positioned as a futuristic replacement for government work; it’s being positioned as a kind of accelerated legal librarian—one that can sift through large volumes of text quickly enough to make modernization feel less like a multi-year slog and more like an ongoing process.
That framing matters, especially because Hochul has also recently signed a moratorium on new AI data centers in New York. The juxtaposition—using AI internally while pausing certain kinds of AI infrastructure expansion—captures a tension that’s increasingly common across governments: the desire to benefit from AI’s productivity gains, paired with caution about the costs, risks, and externalities that come with scaling it.
In the podcast conversation, Hochul offered examples of the kinds of antiquated laws her administration is looking to surface. Among them were a $25 fee required to take a dog hunting, and a requirement that pregnant people need a permit to work after midnight. These aren’t just quirky trivia; they’re the sort of provisions that can become symbolic of a broader problem—rules that remain on the books long after the social or administrative logic that produced them has faded.
What makes Hochul’s approach notable is the scope implied by her phrasing. “Every single rule, regulation, [and] policy” suggests a comprehensive review rather than a targeted audit of a few high-profile areas. That’s a big deal because legal modernization often fails not due to lack of will, but due to the sheer volume of material involved. Even when lawmakers and agencies agree that something should be updated, the practical work of locating, interpreting, cross-referencing, and deciding what to change can take years. Hochul acknowledged that point directly, saying that reviewing all of the laws “probably would have taken five years at the staff level” without automation.
That estimate is telling. It implies that the bottleneck isn’t necessarily the final decision-making—what to repeal, amend, or clarify—but the preliminary work: finding what exists, understanding how it fits into the rest of the legal ecosystem, and flagging what might be obsolete. AI, in this context, is being used to compress the discovery phase. It can read quickly, categorize, and surface patterns that humans might miss simply because the material is too large or too fragmented across statutes, regulations, and policy documents.
Still, there’s an important nuance in how such systems are typically used. When officials say AI is “analyzing” rules, it doesn’t automatically mean the AI is making legal determinations on its own. More often, the technology functions as a first-pass filter—highlighting candidates for review, suggesting where contradictions or outdated language might exist, and helping staff prioritize what deserves human attention. In a government setting, that distinction is crucial: even if AI can accelerate the search, the authority to interpret and change law remains with people.
Hochul’s remarks also land in a moment when AI governance is becoming a public issue rather than a behind-the-scenes technical debate. The moratorium on new AI data centers signals that New York is not treating AI deployment as a purely private-sector matter. Data centers are physical infrastructure with environmental, economic, and grid-related implications. They also raise questions about transparency and accountability: who benefits, who bears the costs, and how communities are consulted.
So when Hochul says her administration is using AI to clean up the legal code, it raises a question that many observers will be asking: if the state is pausing certain AI expansions, what exactly does it consider acceptable use of AI within government? The answer likely lies in the difference between internal administrative tooling and large-scale infrastructure buildouts. But the political optics are still complicated. Critics may argue that the state is trying to have it both ways—embracing AI’s benefits while restricting its growth. Supporters may respond that the moratorium is about responsible planning, not rejection, and that internal use is part of building the capacity to regulate and understand AI better.
There’s another layer here: legal modernization is one of the most natural places to apply AI, because law is text-heavy and structured around language. Statutes and regulations are essentially long-form documents with defined terms, exceptions, and cross-references. That makes them well-suited to computational analysis—at least in the sense that AI can parse and compare language at scale. If the goal is to identify outdated requirements, AI can help by detecting anomalies: provisions that appear inconsistent with current practice, language that references obsolete agencies or procedures, or rules that no longer align with modern regulatory frameworks.
But “outdated” is not always straightforward. A law can be old without being wrong. It can be archaic in wording while still serving a purpose. It can be politically sensitive to remove even if it seems redundant. That’s why the most valuable role for AI in legal review is often not to declare a rule dead, but to generate a shortlist of candidates for deeper human evaluation.
Hochul’s examples illustrate the kind of candidate that tends to stand out. A $25 fee tied to dog hunting might be outdated due to changes in licensing systems, enforcement priorities, or administrative structures. A permit requirement for pregnant people to work after midnight is the kind of provision that would likely trigger immediate scrutiny—not only because it may be obsolete, but because it raises serious questions about fairness and discrimination. Whether such rules are truly enforceable today, how they interact with other labor protections, and what legal precedents exist would still require careful review. AI can help locate and surface these items; it can’t replace the legal reasoning needed to determine what should happen next.
The governor’s comment about saving “five years” also points to a broader trend: governments are increasingly under pressure to modernize faster, but they often lack the staffing and time to do so. AI becomes attractive not because it’s glamorous, but because it can reduce the time spent on repetitive reading and sorting. In a state with thousands of rules and regulations, the difference between “we’ll get to it eventually” and “we can start now” can depend on whether the initial triage is feasible.
That said, there are risks and limitations that any administration using AI for legal review would need to manage carefully. One risk is overreliance on the system’s outputs. If staff treat AI-flagged items as presumptively outdated, they could miss context or misinterpret the significance of a provision. Another risk is bias in the underlying models or training data—especially if the AI is more confident about certain types of language patterns than others. A third risk is transparency: if the public later asks why a particular rule was flagged or deprioritized, the administration needs a defensible explanation that doesn’t rely solely on “the AI said so.”
Even if Hochul’s team is using AI responsibly, the public conversation will inevitably focus on accountability. That’s particularly true because the same technology that can help clean up laws can also be used to automate decisions in ways that affect rights and opportunities. The more AI touches government processes, the more the public will demand clarity about how it’s used, what safeguards exist, and how errors are corrected.
This is where Hochul’s dual stance—AI moratorium on data centers, AI use for legal cleanup—could be interpreted as an attempt to draw a line between different categories of AI activity. Infrastructure expansion has external impacts and requires planning; internal administrative analysis may be seen as lower-risk and more controllable. Yet the line is not always clear to the public. People may wonder whether the state is building AI capacity while limiting AI growth, or whether the moratorium is primarily about leverage and negotiation rather than principle.
Regardless of interpretation, Hochul’s comments highlight a practical reality: governments are already using AI, and the question is increasingly not whether AI will be used, but how it will be governed. Legal code cleanup is one of the more defensible uses because it aims to improve clarity and relevance. Outdated rules can create confusion, impose unnecessary burdens, and sometimes perpetuate inequities. If AI helps identify those problems faster, it could lead to tangible improvements in how residents experience government.
At the same time, the examples Hochul cited underscore why legal modernization is not merely technical. A rule about a fee for dog hunting might be a matter of administrative simplification. A rule that appears to single out pregnant people for special permitting requirements is a matter of rights and equal treatment. That means the AI-driven review could surface issues that are not just “old” but potentially harmful. If so, the stakes rise from efficiency to justice.
There’s also a strategic angle to consider. By using AI to analyze the legal code, the administration may be building institutional knowledge about how laws are structured and how they change over time. That knowledge could inform future legislative drafting, regulatory reform, and even how agencies communicate requirements to the public. In the best-case scenario, AI-assisted review becomes a feedback loop: the state identifies outdated provisions, updates them, and then uses the updated corpus to improve future analysis.
In the worst-case scenario, it becomes a one-off effort that produces a list of recommendations without sustained follow-through. Legal reform is notoriously slow, and even when outdated rules are identified, political and procedural hurdles can delay action. The real test will be whether Hochul’s administration turns AI findings into actual legislative proposals, regulatory amendments, and enforcement changes.
Another question that will likely emerge is how the state will handle the public-facing side of this work. If AI is used to analyze every rule and regulation, residents may reasonably ask: will the