What happened

Reports from former employees and internal documentation revealed that the AI-generated lessons  at the costly and much hyped Alpha School in Austin, Texas, frequently produced faulty lesson plans, "hallucinated" facts, contained ambiguous wording, illogical multiple-choice answers and errors that fail to meet standardised test expectations. 

These technical failures treated students as "guinea pigs" in an unregulated educational experiment, undermining students’ learning and confidence. Former employees describe feeling anxious and betrayed when assessments do not logically match the questions, because they could not trust whether their exams are valid measures of their knowledge.

Students were also forced to hit rigid, AI-set goals without sufficient human support, leading to overwork and burnout, and their performnace was tracked using surveillance techniques, creating a high-pressure environment.

Allegations also surfaced that the school’s AI was trained by scraping data from other online courses without permission, undermining intellectual property standards.

Why it happened

The core problem is the decision to build a school model that replaces much human teaching and quality control with generative AI, while treating pupils and their data as training material to refine the system. 

Leaders promoted the idea that “all educational content is obsolete” and that AI could fully personalise learning, creating pressure to automate lesson generation and review rather than invest in robust human oversight.

 Internal notes reportedly show that staff knew the AI had a significant hallucination/error rate yet still used AI to check AI-generated lessons, meaning flawed outputs were allowed to correct themselves instead of being systematically reviewed by qualified educators. 

Transparency to parents and students appears limited: they were sold high test scores and “2‑hour school days”, while internal accounts suggest many students needed more time and human tutoring to compensate for AI shortcomings, raising questions about misleading marketing. 

The alleged scraping and reuse of other platforms’ educational content without permission, and at least one external platform cutting ties, indicate weak governance around data sourcing, intellectual property, and compliance.