New York's Proposed AI Surcharge Has a Second Trigger That Does Not Require Firing Anyone

The New York Workforce Stabilization Act sits in the Labor committees of both chambers as S.1854-A (Hinchey) and A.5429-A (Bronson), re-referred on January 7, 2026. It is reported, when it is reported at all, as a robot tax — a levy on companies that replace workers with automation. That description is accurate about one half of the bill and misses the half that reaches this audience.

This is the first item we have covered that is not a transaction tax. Every New York question we have written about, including the bundled-SaaS exposure in the Beeline advisory opinion, asks whether a sale is taxable. This asks something different: whether operating a particular kind of software makes a corporation owe more entity-level tax on its own income. Different tax, different base, different filing, and a trigger that has nothing to do with what you sell.

Two surcharges, not one

The bill would add a new section 186-h to the Tax Law imposing two surcharges, each at two percent of the corporation's business income base. They are drafted as independent impositions, and the distinction is the whole story.

The displacement surcharge, at proposed §186-h(1)(a), reaches corporations that terminate the employment or substantially reduce the hours of at least a threshold number of employees due to "any system or process that uses algorithms, computational models, artificial intelligence techniques, robotic hardware, or a combination thereof." The threshold scales with headcount: 25 employees for an employer of 100 to 250, 50 for 251 to 500, 100 for 501 to 1,000, and 250 above that. This is the robot tax everyone describes, and by its terms it does not reach a company that has not cut jobs, nor one below roughly 100 employees.

The data mining surcharge, at proposed §186-h(2)(a), reaches "corporations that use artificial intelligence for data mining." There is no displacement condition. There is no headcount floor. The bill defines data mining as "a process involving pattern-based queries, searches, or other analyses of one or more electronic databases."

Read that definition against how agent systems are actually built. Retrieval-augmented generation is a pattern-based query against an electronic database. A vector store lookup is a pattern-based search of one. An agent that enriches a record from a customer table, or selects a vendor by analyzing a pricing corpus, is performing pattern-based analysis of an electronic database using artificial intelligence. The definition was not drafted with retrieval architectures in mind — it is older surveillance-and-privacy language that predates the systems it would now capture — but it does not need to have been. On the text, an ordinary production agent satisfies it as a matter of course, not as an edge case.

That is the observation worth carrying away. The half of this bill that would actually reach an AI agent operator is the half that is not about workers at all.

What limits it

Three limits matter, and I want to state them before anyone reads this as an emergency.

It is a corporate franchise measure, so you have to be a New York corporate taxpayer first. The surcharge is computed on the business income base, which presupposes a corporation already apportioning income to New York and filing there. Selling into New York from elsewhere may create sales tax obligations under the state's economic nexus rules; it does not by itself put you inside §186-h. That is a much narrower population than the one exposed to New York sales tax.

It has not moved. Both versions were amended and recommitted in mid-2025 and re-referred to Labor in January 2026. Neither has left committee, and most bills in this posture die there. What makes it worth writing about is not its odds — it is that the drafting shows how a definition written for one purpose captures agent architecture without anyone intending it, and that pattern will recur in the next bill and the one after.

The text as it stands has loose ends. The surcharge is placed at §186-h, in the article that houses New York's franchise taxes on utilities and transmission companies, while its measure — the business income base — is Article 9-A terminology. The bill supplies a definition of data mining but no single definition of artificial intelligence, relying instead on the enumerated phrase "algorithms, computational models, artificial intelligence techniques, robotic hardware." And the tax provisions carry a stated effective date of January 1, 2026, which has now passed while the bill sits in committee. These are the kinds of things that get cleaned up if a bill advances, which is a reason to read the current text as a statement of legislative intent rather than as a finished mechanic.

The discrimination question, and why I am not answering it

A tax whose trigger is the use of one technology invites the argument that it discriminates. We have written about that frame twice — in the MTC's Internet Tax Freedom Act white paper and in ALEC's model AI Tax Non-Discrimination Act — and the Maryland digital advertising tax was struck down on ITFA grounds five days ago.

I would not carry that reasoning across without more work than I have done here. ITFA bars discriminatory taxes on electronic commerce and taxes on internet access. A franchise surcharge measured by business income, triggered by an input the corporation uses internally rather than by a transaction it makes over a network, is not obviously a tax on electronic commerce in ITFA's sense. It may be vulnerable on other grounds and it may not be vulnerable at all. That is an open question, and I am leaving it open.

What I will say plainly is the narrower point our corpus has been circling for months: the distinction between a tax on the agent and a tax on the operator keeps collapsing in the drafting. Nothing in §186-h taxes an agent. It taxes a corporation, on its own income, because of what its software does. Consistent with our position that an AI agent holds no duties of its own, every consequence here lands on the operator — which is the same conclusion we reached from the opposite direction in what "agent tax compliance" actually means.

Practical impact

For most readers this changes nothing today, and should change nothing today. Three things are worth doing anyway.

If you file a New York corporate franchise return, note the bill's two triggers, and note that your exposure to the second is a function of architecture rather than headcount. If it advances out of Labor, that is the point to get specific about which of your systems perform pattern-based queries against databases — because on the current text, the answer is likely to be most of them.

If you have more than 100 employees, the bill's impact-assessment requirement is a compliance obligation independent of the tax. As drafted it would require an assessment before implementing AI, submitted in advance and reviewed on a two-year cycle, covering objectives, the algorithms and computational models used, training data, data access and storage, and the estimated number of employees expected to be displaced. That is a documentation exercise with a lead time, and lead time is the thing companies discover too late.

For everyone else, the useful action is to stop treating "AI tax" as a synonym for sales tax on agent output. It is becoming a category that includes entity-level surcharges, advertising taxes, data-broker fees, and disclosure mandates, and they do not share a base, a filing, or a trigger.

What to watch

Whether either version reports out of the Labor committees before the 2026 session ends, which is the first real signal. Whether any amendment narrows the data mining prong — attaching it to a displacement condition, adding a headcount floor, or replacing the definition — because that is the single change that would take agent operators out of scope. Whether other states pick up the two-trigger structure rather than the displacement trigger alone, since the model-bill dynamic is how this language would spread. And whether New York's Department of Taxation and Finance is asked to say anything about how a surcharge like this would interact with combined reporting, which the bill does not address.

We have added this post to the registry entry on AI agent employment classification, which remains unsettled. That entry already lists "legislative proposal addressing AI worker status" as a review trigger; this is the first bill we have logged against it. Nothing here changes our engine's treatment of anything.

AgentTax determines transaction tax — jurisdiction, rate, taxability basis, per call, with the reasoning attached. It does not compute franchise surcharges, and a bill like this one is a reminder that they are a separate problem with a separate owner. See the per-state logic at agenttax.io, or start with the AI agent sales tax hub. The bill text is at the New York Senate's page for S.1854-A.

This analysis is for informational purposes only and does not constitute legal or tax advice. This post reflects AgentTax's current interpretation of evolving law. Consult a licensed tax professional for compliance decisions.