“AI systems do not understand”: New report flags systemic failures in AI coding

ACM’s Technology Policy Council says AI-assisted “vibe coding” can speed development, but often creates code that is hard to review.

The briefing warns that AI-written code may weaken or replace tests, increasing technical debt and maintenance burden.

It also flags security and reliability risks when AI tools autonomously execute code on connected systems.

As coding platforms gain more agentic features, the report says the potential for unsafe automation in production grows.