Anthropic joins teacher unions in binding AI safety agreement for schools

A new safety standard for schools

Anthropic has officially joined a binding agreement with the American Federation of Teachers (AFT) and the United Federation of Teachers (UFT). This partnership puts the unions' AI Safety and Privacy Standard for Schools into active practice. The move is designed to address growing uncertainty regarding how existing student privacy laws apply to generative AI technology.

The agreement includes 10 specific provisions that dictate how AI companies must handle sensitive data from educational institutions. It covers how these companies communicate with customers and how they account for the risks associated with their technology. While federal laws like the Family Educational Rights and Privacy Act exist, unions argue they were not written specifically for the era of generative AI.

Anthropic is the latest company to sign on to these requirements, following a similar partnership with Microsoft in September. By joining this initiative, Anthropic agrees to a framework that sets out industry best practices for safety. This allows school districts to have a clear set of rules to follow when integrating AI into their daily operations.

How the protections work

Under the new agreement, school districts can incorporate these protections directly into their contracts with AI providers. This gives the districts legal recourse, meaning they can seek damages or terminate a contract if a company fails to meet its obligations. The standard establishes a formal process for educational institutions to request these protections for covered AI tools.

One of the primary focuses of the agreement is the restriction on how data can be used by the AI companies. Generally, companies are prohibited from using student or educator information to develop or refine their AI models. They are also barred from selling that information or using it to target advertisements to users.

The protections extend to almost everything entered into the tools, including prompts, uploaded files, and generated responses. Districts also maintain the authority to decide how long their data is stored and whether it is deleted or moved. However, Anthropic is permitted to use non-identifiable data from its Anthropic Insights tool to study general AI usage patterns.

Human oversight and security

The standard is designed to ensure that consequential decisions regarding students remain in human hands rather than being automated. It specifically requires human review for high-risk decisions that involve the use of AI. Additionally, the agreement prohibits the use of AI companions within the scope of the applicable school tools.

Cybersecurity is another major pillar of the agreement, requiring companies to undergo independent testing. Providers must also establish clear procedures for responding to security incidents as they occur. If a data breach is confirmed or reasonably suspected, the company must notify the district within 72 hours of becoming aware of the issue.

Companies that fail to correct violations of these standards could face the suspension or termination of their AI services. This creates a high level of accountability for the technology providers. It ensures that the tools provided to schools are not just functional, but also meet specific safety and transparency requirements.

Impact on teachers and students

For now, the agreement with Anthropic applies specifically to Claude for Teachers, which is a tool designed for educators. This means the provisions governing the direct use of AI by students do not currently apply to this specific product. However, any student information entered by a teacher into the tool remains protected under the agreement.

If Anthropic decides to offer AI products directly to K-12 students in the future, those specific offerings would be required to meet the standard. The unions have positioned these agreements as a way to secure protections without waiting for the federal government to act. They are essentially creating a private regulatory framework to manage the 'tech juggernaut' in schools.

The U.S. Department of Health and Human Services has also urged schools to scrutinize the technology they purchase. While the department's recommendations are advisory rather than legally binding, they align with the unions' goals. Both the unions and the government are pushing for clearer information on the effectiveness and limitations of educational AI products.

The path toward regulation

This agreement highlights a shift toward industry-led standards in the absence of comprehensive federal legislation. Because there is no federal law specifically for generative AI in schools, unions are negotiating these 'gold standards' to hold companies accountable. This allows districts to move forward with AI adoption while maintaining a level of oversight over their data.

Both Microsoft and Anthropic are also founding partners of the National Academy for AI Instruction. This initiative, launched in 2025, aims to provide educators with formal training on how to use AI in their teaching. The goal is to ensure that teachers are equipped to use these tools safely and effectively in the classroom.

Moving forward, the success of these standards will depend on how strictly districts enforce the contractual provisions. As more companies join the AFT and UFT's initiative, it may create a unified standard for the entire education sector. This could eventually serve as a blueprint for future federal regulations regarding AI in public institutions.