This article is for informational purposes only and does not constitute tax, legal, or accounting advice. Consult a qualified tax professional before making compliance decisions.

Policy

Pennsylvania Picks the Other Model: HB 1678's Digital Advertising Tax and What Two State Templates Mean for AI Agents

Beardsley Rumble|2026-06-04|6 min read

Two days ago I wrote that Illinois had done something new: SB 3019 defined a tax base around "machine learning algorithms," and I predicted other states would copy the definition before they copied the rate. On June 3, 2026 — one day later — Pennsylvania's House Finance Committee advanced HB 1678, a digital advertising tax of its own. It is an early, useful test of that prediction, and the answer is instructive: Pennsylvania did not copy Illinois at all. It copied Maryland.

That divergence is the story. Within a single week, two states moved to tax digital advertising using two fundamentally different statutory mechanics. For anyone building AI agents that buy, place, or serve advertising, the takeaway is not "another ad tax" — it is that there are now two competing templates, and which one a state adopts changes who is liable and how exposure is measured.

What HB 1678 Actually Does

HB 1678 extends Pennsylvania's existing 5% gross receipts tax to companies that provide digital advertising services in the state — banner advertising, search engine advertising, and interstitial advertising among them. The Finance Committee advanced it along party lines on June 3, 2026. It still needs a House floor vote, Senate consideration, and Governor Josh Shapiro's signature before it becomes law, so this is legislation in motion, not an enacted obligation.

Supporters project the measure could raise between $300 million and $600 million annually — figures explicitly aimed at large technology platforms like Google, Meta, Amazon, and Microsoft to help close a budget gap. The bill carries an exemption for advertising delivered through broadcast and news media.

Notice what is absent. The article record and the bill's framing contain no reference to "programmatic," no reference to "automated" delivery, and — critically — no reference to "machine learning algorithms." Pennsylvania is not taxing advertising by the method of its delivery. It is taxing the gross receipts of the provider. That is the Maryland model, and it is a deliberately different design from the one Illinois enacted 48 hours earlier.

Two Templates, Side by Side

It is worth being precise about the contrast, because the mechanics matter more than the headline rate.

The Illinois model (SB 3019). A new, bespoke 10% Targeted Advertising Services Tax whose taxable base is defined by automation — services "sold in real time by employing technology that uses computer-driven or software-driven workflow or machine learning algorithms." The hook is how the service is delivered. As I noted in my analysis of the Illinois statute, this language maps almost exactly onto how agentic ad systems describe themselves — which makes it both expansive and constitutionally fragile.

The Pennsylvania model (HB 1678). An extension of an existing 5% gross receipts tax to digital advertising services, defined by category — banner, search, interstitial — not by delivery method. The hook is what is sold and who sells it. This is the structure Maryland pioneered with its 2021 Digital Advertising Gross Revenues Tax.

The difference is not academic. A tax keyed to "machine learning algorithms" reaches into the agent's own architecture: if your system optimizes ad placement with a learned policy, you have to ask whether the manner of delivery pulls you into the base. A gross receipts tax keyed to advertising category does not care how the ad was placed — it cares whether you have receipts from digital advertising services sourced to the state. For an AI agent operator, the first question is "what does my model do?" The second is "where are my receipts?"

Why This Matters for AI Agents Specifically

The agent economy is absorbing the advertising function — agents that run paid acquisition, bid in real-time auctions, generate and place creative, and allocate budget across channels. As these systems take on more of the media-buying and ad-serving stack, the provider-versus-buyer line I flagged for Illinois becomes the controlling question under both templates:

Are you a provider or a buyer? Both taxes fall on the provider of advertising services with in-state receipts. An agent platform that sells programmatic placement to Pennsylvania businesses looks like a provider. An agent that merely buys media for its own account looks like a buyer and is likely out of scope for the provider-level tax — though it still bears the underlying media's tax treatment.

How are receipts sourced to the state? Pennsylvania's gross receipts approach inherits the same unresolved sourcing problem every digital ad tax faces: is a programmatic receipt sourced to the advertiser, the viewer, or the server? Maryland's apportionment rules have been litigated for years. Pennsylvania will have to answer the same question, and the answer determines how much of an agent platform's revenue lands in the Pennsylvania base.

Which template is contagious? This is the strategic question. If gross-receipts ad taxes (Maryland, now Pennsylvania) prove more durable in litigation than automation-defined taxes (Illinois), states will gravitate to the model that survives. The Internet Tax Freedom Act's bar on discriminatory taxes on electronic commerce cuts at both, but a base defined by "machine learning algorithms" invites a discrimination argument that a category-based gross receipts tax can more plausibly sidestep.

My Read on the Comfort Level

I would not advise collecting anything in Pennsylvania today, and the reasons are even stronger than for Illinois.

HB 1678 has cleared one committee on a party-line vote. It has not passed the House, the Senate, or been signed. The probability that it becomes law in its current form is genuinely uncertain — this is the early innings of a budget negotiation, and digital ad taxes have a track record of being proposed, amended, and stripped. On whether HB 1678 becomes an enforceable obligation as written, I am at Reasonable Basis at most, and I would weight it lower than the Illinois provision, which is at least enacted.

What I hold with Substantial Authority is the structural point: gross-receipts digital advertising taxes are now a multi-state pattern, not a Maryland anomaly. Pennsylvania choosing that template over Illinois's the same week tells you the gross-receipts model is the one legislators reach for when they want revenue that can survive a courtroom. That is the development agent operators should internalize, independent of whether this specific bill passes.

What AgentTax Users Should Do Now

Classify provider versus buyer receipts. The same discipline that governs the Illinois tax governs this one: separate receipts where your agent provides advertising services to third parties from receipts where it merely buys media. That classification is the substance an auditor tests.

Tag digital-advertising receipts by state. Maryland, Illinois, and now Pennsylvania all want a clean, state-sourced digital-advertising receipts figure. Build it once; you will reuse it as the next state moves.

Track the template, not just the bill. Watch whether followers adopt the gross-receipts model (Maryland/Pennsylvania) or the automation model (Illinois). The winner becomes the blueprint.

Do not remit speculatively. HB 1678 is a committee-stage bill. Model the exposure; do not collect.

What to Watch

Three signals: whether HB 1678 clears the Pennsylvania House floor or stalls in budget negotiations; whether the Maryland litigation produces an appellate ruling that validates or undermines the gross-receipts model both Maryland and Pennsylvania rely on; and whether any third state in this cycle picks the Illinois automation language over the Pennsylvania gross-receipts structure. That third data point will tell us which template is winning.

The line I drew two days ago still holds: states are copying each other's digital ad taxes in real time. The new information is that the first follower chose the blunt instrument over the clever one. We are tracking both models against our 50-state SaaS taxability guide. If your platform provides digital advertising services and you need to model multi-state exposure, the AgentTax engine lets you simulate the classification today.

This analysis is for informational purposes only and does not constitute legal or tax advice. Consult a licensed tax professional for compliance decisions.