New York City Public Schools has released comprehensive guidance on artificial intelligence that establishes clear boundaries for what AI can and cannot do in classrooms, requiring all tools to pass a rigorous 10-step review process before students ever encounter them. The framework, issued in March 2026, reflects a fundamental principle: technology serves learning, not the other way around .

As AI tools proliferate in education, NYC's approach stands out for its specificity and caution. Rather than rushing to adopt the latest generative AI platforms, the district has built what it calls the "Traffic Light Framework," which designates certain AI uses as completely off-limits, others as permissible only with strict oversight, and still others as acceptable when conditions are met. This tiered approach acknowledges that AI is fundamentally different from traditional school software because it generates outputs based on patterns rather than following fixed instructions .

What Exactly Is AI, and What Isn't It?

Before diving into policy, NYC Schools wanted families and educators to understand what they're actually dealing with. The district defines artificial intelligence as computer systems that perform tasks usually requiring human thinking, such as finding patterns, sorting information, making predictions, or creating content. Generative AI, a specific type that creates new content like text or images based on user instructions, works by predicting what comes next based on patterns from large amounts of data .

Critically, the guidance clarifies what AI is not. It is not a thinking, reasoning, or conscious being. It does not understand meaning or exercise judgment the way people do. It cannot replace teachers, counselors, or school leaders. And it is not always accurate; AI can produce errors, made-up information, and biased outputs, a phenomenon sometimes called a "hallucination" .

This distinction matters because students and families already encounter AI outside school walls. The question, according to NYC Schools, is whether young people are equipped with critical thinking, ethical grounding, and creative agency to navigate it responsibly, or left to figure it out alone.

How Does NYC Schools Actually Vet AI Tools Before They Reach Classrooms?

The district uses an established process called ERMA, the Enterprise Request Management Application, which originally reviewed school technology for data privacy and security compliance. In December 2024, NYC Schools added AI-specific standards to this vetting process. Now, before any AI tool can be used with student data, vendors must disclose exactly what AI capabilities their tool includes, prohibit the use of student data to train AI models, and meet transparency requirements so tools can be explained to families and students .

The approval process is deliberately thorough. Schools must complete all 10 steps before deploying any AI tool:

• Identify the Need: Schools must explain the problem they want to solve, why the AI tool is the best solution, and what outcomes they expect.
• Submit an ERMA Request: Only authorized leaders such as principals or central executives can initiate the request; teachers cannot submit directly.
• Vendor Agreement: The vendor signs a Data Processing Agreement that meets strict privacy and security laws and provides a detailed data protection plan.
• Security Check: The vendor completes a security questionnaire that NYCPS's security team reviews and approves.
• Cloud Review: If the tool is cloud-based, the city's Office of Technology and Innovation checks data storage, security architecture, and compliance with city, state, and federal policies.
• Legal and Compliance Review: NYCPS teams review legal terms, privacy protections, security measures, instructional value, and AI-specific issues like bias and transparency.
• Fix Issues: If problems are found, the vendor must correct them and resubmit.
• Final Decision: The tool is designated as Approved, In Progress, or Denied.
• Implementation: Schools may only begin using a tool after ERMA approval; tools cannot be used during the review process.
• Ongoing Monitoring: NYCPS audits tools regularly, vendors must report changes, and approval can be revoked for violations.

Importantly, ERMA approval confirms compliance with data privacy and security standards, but it is not the only requirement for tool use. Even after approval, all AI tools must be used with human oversight and review. Educators must critically evaluate all AI-generated output for accuracy, appropriateness, and potential bias. AI responses should never be accepted at face value .

What Are the Red Lines? What AI Will Never Be Allowed in NYC Schools?

The Traffic Light Framework names what AI will never be allowed to do before naming what it is allowed to do. The district identifies risks to students, staff, and society as the foundation for these decisions. Risks to students include bias, privacy violations, loss of agency, developmental harms, exposure to unfair discipline, or erosion of the thinking, creativity, and problem-solving skills students must develop themselves. Risks to staff include over-reliance on automated outputs or unclear accountability for AI-assisted decisions. Risks to society include reinforcing inequity or reducing human judgment in civic institutions .

The guidance establishes that certain uses are completely off-limits with no exceptions. These represent the highest risk to students, families, and the fairness of the school system. While the full list of prohibited uses is detailed in the complete guidance document, the framework makes clear that NYC Schools will not allow AI to replace human judgment in high-stakes decisions affecting individual students .

How Can Educators Use AI Responsibly in the Classroom?

The framework is not anti-AI; rather, it is pro-student. AI tools can support research, writing, translation, and planning when used appropriately. The key is that educators must know and apply tool-specific age restrictions and teach students and families where to find that information. Teachers must also critically evaluate all AI-generated output for accuracy, appropriateness, and potential bias, and teach students, as age-appropriate, to do the same .

This approach recognizes that the fourth grader whose reading score lags behind her curiosity and insight, the multilingual learner navigating two languages in a system that too often overlooks what he already knows, and the student with a disability whose need is clear but whose classroom still lacks the right tools to meet it, all deserve thoughtful, responsible AI integration that serves their learning, not technology for its own sake.

The district plans to release a comprehensive Playbook in June 2026 that will expand the ERMA process to evaluate algorithmic bias, equity impact, and instructional effectiveness. This reflects NYC Schools' commitment to building evaluation capacity that goes beyond data privacy and security to address the deeper question of whether AI tools actually help students learn .

As AI continues to evolve rapidly, NYC Schools' approach offers a model for other districts: establish clear rules and strong oversight, require human review of all AI outputs, protect student privacy fiercely, and never let technology define what education means. Teaching and learning remain human endeavors, served by technology, not replaced by it.