Anthropic, an AI safety and research company that’s working to build reliable, interpretable, and steerable AI systems, is partnering with Pratham, one of the largest education non-profits, to develop AI-powered tools that help students learn more effectively. 

The first product of this partnership is the “Anytime Testing Machine (ATM).” This end-to-end assessment system, powered by Claude, generates curriculum-aligned tests, digitizes handwritten student answers, provides automatic grading, and delivers personalized feedback. This way, students not only learn how they scored but also receive guidance on what to do next. Crucially, ATM is designed to strengthen learning pathways, not replace them. To ensure pedagogical integrity and alignment with classroom realities, human oversight remains central to the use and delivery of ATM, with teachers and community instructors reviewing and contextualizing AI-generated feedback before it is shared with learners.  

The collaboration has been highly iterative and implementation-focused. Teams from Anthropic and Pratham met regularly over several months to jointly refine the system—not only to deploy the model, but to improve question clarity, calibrate grading criteria, and adapt feedback to ensure it is accessible for second-language learners. 

For initial pilots in India, ATM was designed to operate within the country’s systemic and lived realities, not outside them. For instance, the design considered challenges from the context, including assessments that often generate scores without actionable feedback, under-resourced schools, high pupil–teacher ratios, limited individualized support, uneven connectivity, and device access. Through these pilots, ATM has already reached 6000 learners, including school students and older women returning to take their Grade 10 examinations. In 2026, the ATM will be piloted with an additional 15000 learners in the country. 

“AI tools like Claude give us a way to reimagine learning for students who do not have access to advanced educational resources,” said Madhav Chavan, Pratham’s co-founder. “In addition to providing personalized support to understand the textbook, the ATM innovation will help children verify and authenticate their knowledge beyond the textbooks.”  

Chavan sees assessment as just the starting point for something larger: a system where students can eventually be tested on any topic, not only what the curriculum prescribes, and receive a credential for what they actually know. “Instead of asking children questions about the curriculum, ask them what they know,” Chavan said. “If we flip that, we can flip the education system from being a filtration mechanism to one that offers differentiated pathways based on a child’s interests and background knowledge. Before AI, this was not possible.”

The partnership is already expanding. Anthropic will support Pratham’s Tech in TaRL (Teaching at the Right Level) initiative, an AI-powered teacher support system with a randomized controlled trial planned for several thousand students. Pratham and Anthropic are also exploring educational digital public infrastructure (including knowledge graphs), as well as opportunities to address the needs of learners in countries like Kenya, Rwanda, and other communities across the Global South. Over the next three-years, Pratham’s goal is to evolve the ATM into a learning and credentialing engine that recognizes competencies gained through non-linear pathways, available to learners worldwide.