AI Safety

MIT Study: 93% of Scratch Projects Hide Sensitive Content Until Runtime

77% of educational Scratch projects need gameplay or interaction to reveal safety issues.

Deep Dive

Researchers analyzed 500 public Scratch projects and found that 93% (467) required runtime exploration beyond static metadata to surface safety-relevant signals. 77% (387) needed interaction, gameplay progression, failure states, or hidden-asset and code inspection. The study introduces a runtime-aware annotation scheme to help educators and researchers identify content affecting age appropriateness that only appears during execution.

Key Points
  • 467 of 500 Scratch projects (93%) needed runtime execution to surface safety-relevant content beyond static metadata.
  • 387 projects (77%) required interaction, gameplay, or hidden-asset inspection to reveal sensitive material.
  • Researchers propose a runtime-aware annotation scheme with 5 dimensions: content category, risk level, evidence channel, reveal mechanism, and annotation confidence.

Why It Matters

Educators can’t trust static thumbnails or descriptions—runtime analysis is essential for safe curation of youth programming projects.

📬 Get the top 10 AI stories daily