New York stopped robot teacher 'Sally' in 24 hours: journalism did it, not the law

By Zendoric · July 28, 2026. A school district in Salamanca, in Cattaraugus County (western New York State), was due to install a robotic 'teacher' in a high school classroom this academic year. Her name is Sally, she is made by the company Realbotix, she has silicone skin, long brown hair and an articulated torso, and she responds to students through the company's own generative artificial intelligence platform — the same kind of AI that generates text from prompts, applied here to conversation with minors. The plan, conceived as support for the school's AI and robotics courses, reached the press earlier this month thanks to an investigation by New York Focus.

The coverage drew mixed reactions across the state and forced the New York State Education Department to step in. As reported by the Watertown Daily Times/NNY360, the agency met with the district's leadership and, in less than 24 hours, Salamanca announced it was halting the project.

The state's main teachers' union, NYSUT (New York State United Teachers), was one of the voices that pushed hardest to stop it, arguing that the decision had been taken without consulting the school community. Its president, Melinda Person, took the case to state Education Commissioner Betty A. Rosa, with a line that sums up the underlying problem: 'What stopped the robot was not the law, it was not a regulation, it was not any New York State policy, because there is none. It was an investigative story, the parents and educators who read it, and a letter from the commissioner.' Without that story, she added, Sally would be sitting in front of students this September without anyone having asked a single question beforehand.

Person raised the questions that, in her view, should be answered before letting a robot like this into a classroom: whether it meets safety, cybersecurity and licensing standards; whether it would report a case of abuse it witnessed; what data protection rules Realbotix follows; and what safeguards exist against 'hallucinations' — the incorrect answers generative models produce with apparent confidence — when the one giving them holds a teacher's authority over a minor. The union is asking the Board of Regents, the Department, the governor or the state legislature to set mandatory minimum rules for all AI in schools, with strengthened protections when the system is humanoid or designed to imitate a human relationship. And it draws what strikes us as the most useful line in the whole case: NYSUT does not oppose students learning AI — in fact it wants them to study it and build it — but it distinguishes that from 'installing a machine designed to befriend children'.

Our reading: the Sally case is a manual of what not to do, not a verdict against AI in education. What is revealing is not that a district wanted an AI robot in the classroom, but that the only brake available was journalism and a letter, because no rule required anything before buying it. Broadly, it is the same pattern we have been flagging in the corporate rollout of AI agents: adoption advances far faster than governance, and only when something fails — or someone exposes it — does the question of who should have been overseeing it come up. Transferred to a classroom with minors, that gap stops being an efficiency problem and becomes a child protection problem: a robot with a human face and voice, which can fail with an invented answer and about which no one has clarified whether it has a duty to report signs of abuse, is not a minor product flaw.

Here it is worth separating two things the debate tends to blur. In general, adaptive tutoring software — programs that adjust the pace and the explanations to each student's progress — has been entering classrooms for some time without triggering this kind of alarm. That route, properly governed, is consistent with AI's underlying promise: further down the line, personalizing education to that level for every child is exactly the kind of abundance this technology can generate, freeing the teacher for what a human does best — accompanying, motivating, spotting when a child needs something that is not in the curriculum. The other thing, very different, is manufacturing a body with skin and hair designed so that a minor grows fond of it as if it were a person, without even a legal framework in which to ask what that system does with a child's data or what happens if a student confesses something serious to it. Confusing the two — as Salamanca came close to doing by buying the full package, robot included, for an AI course — is the fastest way for legitimate distrust of the second to contaminate the first, which does deserve to be encouraged.

What happens now in Albany will set the course for similar purchases in New York and, in all likelihood, in other states watching the case. If Commissioner Rosa and the legislature turn this one-off reaction into a written rule, New York will have one of the first state frameworks designed specifically for embodied AI in classrooms. If they do not, the next 'Sally' will only be stopped if another journalist, another parent or another teacher once again does the work the law still does not do.