AI + Education Weekly Update
Week of December 15th, 2025
To catch up on another busy week in AI + Education news, check out our weekly updates for our review and analysis of some of the key highlights worth staying informed about.
Week of December 15th, 2025
Guidance on AI and Children
With widespread AI adoption among children, UNICEF's updated guidance provides a rights-based framework to ensure that AI policies and systems uphold, protect, and fulfill children's rights globally by optimizing opportunities and mitigating risks.
Child-centered AI framework: Protection (do no harm), Provision (do good), Participation (include all children)
10 requirements including establishment of regulatory frameworks, data privacy, and safety
Stakeholders should prepare students for future AI developments
State of AI and CS Education
The State of AI + CS report found that national computer science (CS) access has plateaued and AI policy is lagging behind. The report argues that this proves CS is foundational to AI and should be prioritized by policymakers.
AI relies on core CS concepts like logic, algorithms, data structures, and programming.
4 out of 50 states require AI in their CS standards and none require it for graduation.
High schoolers are not receiving anymore CS instruction than before; only states that have passed a CS graduation requirement show growth in CS
Lawmakers are being urged to adopt policies that clarifies CS is fundamental to AI.
Khanmigo for Teachers Expands Access
Microsoft and Khanmigo have expanded the free pilot of Khanmigo for Teachers to 180+ countries and territories, with hopes of shifting focus from prepwork to teaching and engaging students.
Includes standards-aligned lesson plans
No prompting required; 20+ teacher activities
Available in most countries with exceptions due to legal and partner restrictions; 30+ languages
Student access only available for parents and US learners
Wharton: Expert Personas Don’t Improve Accuracy
Wharton research found that assigning expert roles (“You are a science expert”) to GenAI, does not improve accuracy and may even be harmful in the wrong context.
6 AI models tested on graduate-level questions; Expert roles showed no improvement in accuracy; Gemini 2.0 was an exception
Low knowledge personas (“toddler” or “young child”) reduced accuracy
Matching expert personas (physics expert for physics questions) to question types showed no benefit
Mismatched experts sometimes worsened performance
OpenAI: The State of Enterprise AI
OpenAI analyzed 1 million business customers and found key trends in AI usage: companies are using AI to scale and save time, adoption is growing globally, and usage gaps are widening both between employees and between companies.
ChatGPT Enterprise seats grew 9x YoY; workers save 40-60 minutes daily and 75% now complete tasks they previously couldn't perform, including coding and data analysis
Tech grew 11x YoY; International markets like Australia, Brazil, and the Netherlands are growing faster than the U.S., with 143%+ customer increases
Top-performing workers send 6x more messages than median users and report 5x greater time savings; frontier companies generate 2x more messages per employee
Microsoft: Copilot Usage Report
Microsoft analyzed 37.5 million conversations with Copilot between January and September 2025, finding that AI serves different roles depending on when and how people access it.
Desktop: work and technology topics dominate during business hours; Mobile: health, fitness, and advice seeking dominate across all hours
Usage follows predictable rhythms: programming peaks on weekdays, gaming on weekends, and philosophical questions surge late at night
Shift from productivity-focused usage (heavy programming in January) to broader social integration by September
