On April 7, 2026, the Third Circuit Court of Appeals ruled that publishing the full text of copyrighted standards that have been adopted into law is likely fair use. The case was ASTM International v. UpCodes Inc., and it hinged on a question so fundamental it seems absurd that it took this long to answer: can a private organization copyright the law?
ASTM said yes. UpCodes said no. The court said something more nuanced than either party wanted, and in doing so cracked open the economics of every AI plan review tool in the country.
What UpCodes Actually Did
UpCodes is a platform that publishes building codes in a searchable, hyperlinked format. It posted the full text of ASTM standards, the ones that Philadelphia and hundreds of other jurisdictions have incorporated by reference into their building codes. ASTM, which develops and sells those standards, sued for copyright infringement. They lost.
All four fair use factors went against ASTM. Factor one: UpCodes’ purpose was “transformative,” because informing citizens about the law differs fundamentally from distributing industry best practices. Factor two: laws are facts, sitting at the “periphery of copyright’s core protection.” Factor three: full copying was justified, because you cannot partially inform someone about a legal requirement. Factor four: ASTM could not demonstrate significant revenue harm, and the court found “enhanced public access to the law is a clear and significant public benefit.”
The Fifth Circuit has gone further, holding that building codes lose copyright entirely once adopted into law. Between these two circuits, the legal foundation for treating building code text as licensable intellectual property is eroding fast.
Why This Matters for AI Plan Review
Every AI-powered plan review tool does the same thing: it reads building code text, parses construction drawings, and flags violations. AutoReview.AI in Gainesville claims to reduce review time from weeks to 30 minutes. Archistar AI PreCheck reports an 80 percent reduction in resubmissions for Surrey, Canada, handling 65 percent more applications with the same staff. PermitPlace’s 2026 analysis of 741 cities found the national average initial permit review takes 22.9 days, with 18 cities hitting 180 days, San Francisco averaging 60, and Chicago sitting at 92.
Before April 7, every one of those tools operated in a copyright gray zone. To parse the International Building Code against submitted plans, the software needed the IBC text in its training data or inference pipeline. That text was copyrighted by ICC, and the referenced standards within it, fire ratings from ASTM, electrical requirements from NFPA, accessibility standards from ICC A117.1, were each separately copyrighted by their respective standards development organizations, which meant licensing fees stacked up quickly. For a startup running lean, negotiating content licenses from three or four SDOs before writing a single line of inference code was a real barrier to entry.
That barrier just dropped. Not gradually, not through negotiation, but through a federal appellate ruling that reframed the entire question from intellectual property dispute to public access imperative, a distinction that rewrites the cost basis for every company in the space and every municipality considering a pilot.
22.9 Days, 741 Cities, One Bottleneck
Municipal plan review is where permits go to age, sometimes for months at a stretch in departments that haven’t hired a new plan examiner since 2019. PermitPlace analyzed every city in its dataset and found a 2-to-5x gap between published review timelines and actual commercial experience. A jurisdiction that advertises a 14-day review might routinely take 45. Contractors budget for it, homeowners absorb it, and the carrying costs compound silently: construction loan interest, delayed occupancy, subcontractor scheduling conflicts that cascade into change orders.
AI plan review tools attack this bottleneck directly. Los Angeles Mayor Bass signed Executive Directive No. 19 on April 27, mandating a 60-day total permit timeline for affordable housing and procuring AI pre-plan-check tools within 30 days. San Jose is piloting CivCheck for pre-submission review, targeting 62,200 housing units by 2031. Naples, Florida, partnered with Blitz AI and CityView, and Louisville, Kentucky, brought on Govstream.ai.
None of these announcements mention the copyright question. Not one. They should, because the tool they are procuring needs to ingest the very text that was, until five weeks ago, locked behind licensing agreements that no municipal IT department was equipped to negotiate on a pilot program budget.
What Changed for Startups
Consider the economics before and after April 7. Stark contrast.
| Cost Factor | Before Ruling | After Ruling |
|---|---|---|
| Building code text licensing | Negotiated per SDO | Fair use (adopted codes) |
| Referenced standards (ASTM, NFPA) | Separate licenses required | Fair use if adopted into law |
| Training data acquisition | Legal risk for web scraping | Transformative purpose defense |
| Municipal deployment clearance | Copyright indemnification clauses | Reduced liability exposure |
| Geographic coverage expansion | License per jurisdiction adopted code | Uniform fair use argument |
Six AI plan review startups are currently competing for municipal contracts: AutoReview.AI, Archistar, Blitz AI, Permitify (YC 2025), CivCheck, and Zermit AI (launched March 2026). Before the ruling, each needed to solve the code text access problem independently, either through licensing agreements, partnerships with code publishers, or by operating in a legal gray area and hoping nobody sued. After the ruling, the code text becomes a commodity input available to anyone with a web scraper and a purpose that a court would characterize as informing the public about the law. Competition shifts to inference quality, integration depth, and municipal sales execution.
Nobody Wrote the Liability Clause
Here is where the optimism should stop. Hard.
AIA standard form agreements and ConsensusDocs, the two contract templates that govern the vast majority of residential construction projects in the United States, contain zero AI-specific provisions. Zero. No clause addresses who selects the AI plan review tool. No clause allocates liability for autonomous system errors. No clause determines whether an AI-generated compliance report constitutes professional work product. No clause says whether AI failure qualifies as force majeure.
On May 5, 2026, a federal judge ruled that senior lawyers are personally liable for AI-generated legal errors, even when the AI tool was used by a junior associate. Translate that reasoning to construction: if a plan examiner relies on an AI tool that misreads a fire separation requirement, and the building gets permitted with inadequate fire-rated assemblies, does the liability fall on the municipality that procured the tool, the startup that built it, or the plan examiner who clicked “approve”?
Courts will almost certainly treat AI errors as the deploying party’s responsibility. Lexology’s analysis reaches the same conclusion, and no major AI-in-construction case has been litigated yet, a silence that is temporary. Michelman & Robinson flagged the contract gap in late 2025, noting a projected $48 billion AI-in-construction market operating without contractual guardrails.
ASTM’s Best Argument
Building codes cost real money to develop. Millions. ASTM committees convene hundreds of engineers, run physical tests, review field data, and publish updates on multi-year cycles funded entirely by standards sales revenue. If every AI startup can scrape the text for free and build commercial products on it, that funding model erodes, slowly at first, then all at once as open-access tools replace the purchased code volumes that have subsidized committee work for decades. Not immediately, perhaps not even within five years, but the incentive structure shifts.
The Third Circuit acknowledged this concern and found no evidence of significant revenue harm. That finding was based on current data, and if AI plan review tools proliferate and reduce direct code purchases by municipalities, contractors, and architects who previously bought print or digital editions, the revenue harm argument becomes considerably easier to make in future litigation. This ruling is a preliminary injunction denial, not a final judgment on the merits, which means ASTM can still win at trial.
Only 56 percent of construction investors plan increased AI investment, according to RICS’s 2025 report, and 75 percent of firms lack AI capability entirely. Forty-six percent cite skilled personnel as the top barrier. Free access to building code text does not solve any of those problems. Not one. It removes one cost, the licensing fee, but leaves untouched the harder barriers of talent scarcity, integration complexity, and institutional inertia that actually determine whether AI plan review makes it from pilot to production in any given jurisdiction.
What You Should Do
If you are a builder running permitted work in the Third or Fifth Circuit: ask your AI plan review vendor whether they have been licensing building code text or relying on fair use. The answer tells you how seriously they take legal risk, which predicts how seriously they will take liability when their tool misses something.
If you are a municipality evaluating AI plan review procurement: remove copyright indemnification clauses from your RFP requirements, because they are no longer the barrier, and replace them with accuracy benchmarks, error liability allocation, and professional oversight requirements. LA’s Executive Directive 19 is a useful template for timeline targets, but it says nothing about what happens when the AI gets the code wrong.
If you are a homeowner waiting on a permit: this changes nothing for you today. Municipal adoption of AI plan review tools will take 18 to 36 months to scale beyond pilot programs. In the interim, your permit sits in the same queue, reviewed by the same understaffed department, against the same code text that is now slightly easier for a startup to read. The bottleneck was never the copyright; it was the institutions.
Limitations of This Analysis
ASTM v. UpCodes is a preliminary injunction ruling, not a final merits decision. The case continues, and a different outcome at trial would reverse the fair use determination. No data exists on how many AI plan review startups actually licensed building code text versus operated without formal permission, making the true economic impact of the ruling difficult to quantify. Municipal AI deployment data comes primarily from press releases and pilot announcements; independently verified performance metrics are scarce. PermitPlace’s 22.9-day average is a mean across 741 cities, heavily skewed by outliers on both ends; median review time is 14 days. AutoReview.AI’s “30 minutes” claim and Archistar’s “80 percent reduction” are vendor-reported figures without independent audit. The $48 billion market projection for AI in construction is a common consulting estimate without published methodology.