On September 9, 2026, the American Federation of Teachers, the United Federation of Teachers, and Microsoft announced a National AI Safety and Privacy Standard for U.S. schools, described as the first agreement of its kind and reached after months of negotiation. The core of the agreement is that its protections are legally enforceable: districts can incorporate them directly into new or existing Microsoft customer agreements, with no renewal or renegotiation required, making them contractual terms Microsoft can be held to. The standard rests on three priorities. On privacy, student and educator data will not be used to train AI models, sold, or repurposed, and schools retain control over how data is used, retained, and deleted. On safety, AI systems must include human oversight, avoid harmful or manipulative design, and meet security standards. On transparency, companies must give parents and educators clear, plain-language information about how tools work, what data they collect, and what protections apply. According to reporting that reviewed the underlying agreement, a district that determines Microsoft has materially violated the terms can terminate its agreement and pursue damages, and the protections become available to districts on November 1, 2026. The protections are set out in a binding memorandum of agreement between the National Academy for AI Instruction and Microsoft, with open slots for other AI providers to sign. Microsoft said it would extend the agreement to every school district in the country.

Every district that uses Microsoft products, and, as a template, every district negotiating with any AI vendor. The significance is less the specific terms, many of which mirror what careful districts already seek, than the enforceability: this converts a set of privacy principles into contract terms with a remedy attached, so that a violation is a breach a district can act on rather than a broken promise it can only complain about. It directly answers the problem this bulletin flagged last week, when a Utah audit found that signed privacy agreements often go unverified and unenforced; here the enforceability, including the right to seek damages, is built into the agreement itself. The open invitation for other AI providers to sign the same standard is the part worth watching, because it could establish a portable, district-favorable baseline that pressures the rest of the market.

The agreement reduces exposure for districts that adopt it, but only if they actually do, and only if they understand what it does and does not cover. A district that assumes it is protected without incorporating the standard into its Microsoft agreement gets none of the enforceability. Federal and state monitors, and parents, would ask: Has your district actually added the National AI Safety and Privacy Standard to its Microsoft customer agreement, or merely noted that it exists? Do you know which of your other AI vendors have not signed a comparable enforceable standard, and what your remedy is if they misuse student data? Does your team understand the material-violation threshold and how you would document and act on a breach? The exposure is no longer only whether protections exist on paper, but whether a district has taken the step to make them enforceable and knows how to use the remedy.

On September 10, 2026, Governor Gavin Newsom signed Assembly Bill 1159, by Assemblymember Dawn Addis, as one of thirteen laws in a package aimed at protecting children from technology-related harms. AB 1159 prohibits technology companies from using student data to train or develop AI models, and provides that any company that knows its products are used in schools, and whose products are designed or marketed to students, is barred from selling the data it collects or using that information for anything beyond the student's education. This bulletin flagged AB 1159 as it advanced through the legislature earlier this session; it is now law. The measure expands California's landmark 2014 education technology privacy statute, which limited how companies use data from students in prekindergarten through twelfth grade but applied only to companies that "primarily" serve students and were "designed and marketed" for them. That wording created a loophole that let widely used products, such as general-purpose tools that happen to be popular in classrooms, argue they were not covered. AB 1159 closes that gap by reaching any company that knows its products are used in schools, and it extends the same protections to the data of college students. The new rules take effect next year. Unions representing California teachers, nurses, and college professors supported the bill; the technology group TechNet and the California Chamber of Commerce opposed it.

Every education technology company operating in California, and every district that relies on them, along with districts nationally that will feel California's gravitational pull on vendor behavior. Because California is the largest market and its 2014 law became a de facto national standard, a company is unlikely to build one data practice for California students and another for everyone else, which means the no-training and no-selling rules are likely to propagate well beyond the state. For districts, the most important shift is that the law reaches products that previously escaped the old statute by claiming they were not primarily designed for students, the same general-purpose and embedded tools that are hardest to inventory. The extension to college data also signals that student-privacy protection is no longer treated as a K-12-only concern.

For California districts, AB 1159 raises the floor and, with it, the expectation that districts know which vendors are now covered. A district that cannot say whether a widely used tool falls under the expanded law, or whether its vendors use student data to train AI, will struggle to demonstrate compliance once the rules take effect next year. Federal and state monitors would ask: Which of the tools your district uses are now covered by the expanded statute, including general-purpose products that were previously excluded? Have you confirmed, in writing, that your vendors do not use student data to train AI models or sell it, as the law now requires? For districts outside California, the question is whether you are relying on vendor practices that California is now moving to prohibit, and whether your contracts would meet the standard California has set. The law converts a widely recommended contract term into a legal requirement, and the exposure is being unable to show you meet it.

An Associated Press report published September 14, 2026, examined whether the Microsoft privacy standard would spread across the AI industry, and found the answer already dividing. OpenAI and Anthropic both said they were in talks with the American Federation of Teachers to reach their own safety and privacy pacts. OpenAI called the Microsoft agreement an important milestone for AI safety and privacy in schools and said it looked forward to finalizing its own agreement; Anthropic said it was working with the union to shape a gold standard for safety and privacy. AFT President Randi Weingarten said she had contacted Google directly before the announcement, hoping it would become a signatory. But Google, which the report described as the dominant provider of education technology for America's schools, had not said whether it would offer similar protections and did not respond to the Associated Press. The context sharpening the question is recent: in August, Google turned on its Gemini chatbot for many K-12 students using Google Classroom, which the company says is used by 150 million teachers and students worldwide, reversing its prior policy of allowing Gemini only for students 18 and older. Josh Golin of the online safety nonprofit Fairplay warned that the standard will only be effective if the tech giants sign on, and that a high-profile framework risks being misread by schools as a signal that a tool is safe to use.

Every district, because the standard's protections reach only the vendors that actually sign it, and the vendor most students use every day has not. Google Workspace and Classroom are ubiquitous in American schools, which means a district running on Google does not automatically receive the no-training, no-selling, audit, and transparency protections in the Microsoft standard. Google's August decision to enable Gemini for K-12 students, where it had previously been limited to those 18 and older, puts more student interaction with AI on precisely the platform that has made no comparable enforceable commitment. The risk Golin names is concrete: a district could see the Microsoft standard, conclude the privacy problem is being solved, and overlook that its own dominant AI exposure, Google, sits outside it. The standard is a meaningful floor, but only for the companies that adopt it.

The exposure is a false sense of coverage. The Microsoft standard binds Microsoft, and, if they follow through, OpenAI and Anthropic, but not Google or the thousands of smaller AI vendors districts use. A district that treats one vendor's strong commitment as protection across its whole AI footprint is exposed exactly where it is largest. Federal and state monitors would ask: For the AI tools your students use most, especially Google Gemini through Classroom, what enforceable privacy commitments actually exist? Have you confirmed whether Gemini is turned on for your students, and at what ages, given Google's August change? Are you using the Microsoft standard as a floor to demand from every vendor, or mistaking it for coverage you do not have? The gap between the tools that have signed and the tools students actually use is the exposure.