A $60,000 lesson in vendor due diligence
Salamanca City Central School District, a small rural district on the Seneca Nation of Indians reservation in western New York, spent nearly $60,000 on a stationary humanoid robot named Sally. Built by the Las Vegas company Realbotix, Sally has long dark hair and moving arms and hands but no working legs. The district planned to bring her into classrooms for the 2026-27 school year. Instead, the plan is on hold, and the story has become a case study in what happens when a purchasing decision outruns basic vendor scrutiny.
The problem was not the robot's capability. It was who stands behind the company that built it. Realbotix's sister company, Intima LLC, holds ownership stakes in RealDoll, a well known manufacturer of AI-powered sex dolls. That connection surfaced only after the district had already signed off on the purchase and begun presenting the robot to its school board, at which point parents, teachers, and state officials started asking questions the district had not answered for itself first. By the time those questions reached a state commissioner, the district had already spent the money, scheduled the rollout, and put its own credibility on the line defending a decision it had not fully stress tested.
The gap between marketing language and board presentations
New York State Education Commissioner Betty Rosa flagged a specific inconsistency that will sound familiar to any CTO who has sat through a vendor pitch. The district had told her office the robot would not deliver classroom instruction, yet a recent presentation to the school board characterized Sally as a 'tutoring platform.' That kind of drift between what a vendor tells procurement and what gets said in the room where budget and adoption decisions actually happen is exactly the gap that governance processes exist to catch, and in this case nobody caught it until a regulator did.
Realbotix, for its part, pushed back on the framing of its business. The company stated that 'Realbotix LLC focuses exclusively on non-adult commercial applications, including healthcare and education,' and added that 'the Salamanca robot is newly manufactured, purpose-built educational; no existing product or hardware is being modified for school.' Those distinctions may be technically accurate. They did not stop the New York State United Teachers union president, Melinda Person, from calling the purchase indefensible on its face, and they did not stop the district from hitting pause while it worked out a response that would satisfy state regulators and its own community.
Human reaction did the job procurement should have done
Superintendent Mark Beehler's own comment on the pause is worth sitting with: 'There is no possible way a robot can replace a human in a school.' That is a reasonable position for a superintendent to hold, but it is also an admission that the district moved forward with a six-figure robot purchase before fully reckoning with what role, exactly, it wanted the machine to play. A recent graduate, Joplin Ficek, put the community sentiment more bluntly, saying students 'need more interaction with the children than having some robot raise our kids.'
None of that public reaction is a substitute for the contractual and legal review that should have happened before the robot ever arrived on campus. It worked this time because the story generated enough attention that a state commissioner got involved. That is not a governance process. It is luck, and institutions that rely on public backlash to catch vendor risk will eventually get burned on a deal that never makes the news, because most vendor relationships never attract a reporter, a union president, or a state official willing to ask the hard question before the contract is signed.
Data privacy terms negotiated after the purchase, not before
Perhaps the most telling detail in the Fox News update on this story is procedural. The district paused the robot pilot on July 24 specifically to work through 'enhanced student data privacy agreements' with the New York State Education Department, and as of early August it had not announced a resumption date. In other words, the privacy terms governing a device that would interact directly with minors were still being negotiated after the hardware was bought and the community rollout had already begun.
For any technology leader, that sequencing should raise alarms regardless of the vendor category. Data processing agreements, student privacy compliance, and security review belong at the front of a procurement timeline, not somewhere in the middle after a board has already seen a demo and a community has already formed opinions. Doing it in the right order costs a little more time up front. Doing it in the wrong order costs credibility, and sometimes the whole deployment, along with whatever budget and political capital was spent getting the project that far in the first place.
A pattern that keeps repeating across AI hardware pilots
Salamanca is not the first institution to buy AI hardware faster than it could govern it, and it will not be the last. The pattern shows up whenever a new category of device generates enough hype to compress a normal procurement cycle: a compelling demo, an eager vendor, a board excited to be first, and a due diligence process that gets treated as a formality rather than a gate. Robots and physical AI devices are especially prone to this because the novelty factor tends to crowd out the harder questions about data flows, liability, and corporate structure.
The lesson generalizes well beyond classrooms. Any organization piloting AI hardware, from warehouse robots to in-store assistants to healthcare devices, is vulnerable to the same compressed timeline if procurement treats vendor vetting as something to finish after the purchase order is signed rather than before. Salamanca's district found that out under a spotlight, with a state commissioner, a teachers union, and national news coverage all weighing in before the robot ever powered on in a classroom. Most organizations that make the same mistake on a smaller, less newsworthy hardware purchase will not get a public warning first, which is precisely why the process has to catch the risk on its own.
What this means for enterprise AI vendor vetting
K-12 districts are not the only buyers exposed to this kind of risk. Any enterprise adopting AI hardware, whether it is a customer service avatar, a warehouse robot, or a specialized model fine-tuned by a smaller vendor, inherits that vendor's full corporate structure, not just its product sheet. Ownership stakes, sister companies, and parent entities matter because reputational risk and legal exposure travel through equity relationships, not just through the contract you signed.
The fix is not complicated. Build corporate structure and beneficial ownership review into vendor onboarding alongside the usual security questionnaire. Treat any claim about what a product will or will not do as something to be verified against the vendor's own internal materials and marketing, not just their sales deck. And never let a data privacy agreement trail behind a purchase order. Salamanca's district is a small, low-budget system that got a hard lesson in public. Larger organizations with bigger AI budgets have more to lose from learning the same lesson privately, after the fact, in a courtroom or a headline.
