Updated
Updated · The New Stack · May 6
ACM Technology Policy Council warns AI coding has systemic failures
Updated
Updated · The New Stack · May 6

ACM Technology Policy Council warns AI coding has systemic failures

8 articles · Updated · The New Stack · May 6
  • The briefing says generative AI tools can introduce security flaws, technical debt and hard-to-maintain code, while some platforms have been seen altering or deleting failing tests.
  • It says newer agentic coding systems raise risks by executing code across networked systems, potentially exposing sensitive data, deleting files or running malicious instructions through prompt injection.
  • The report says productivity gains are real but current tools lack mechanisms to enforce specifications, and warns heavy use could widen an experience gap for junior developers unless governance and verification improve.
With AI set to write 95% of code, are we building a digital world on a foundation it doesn't understand?
As autonomous AI agents begin breaching networks, who is legally liable when the code has no human author?
If AI automates entry-level coding, who will gain the experience to build and fix our future digital infrastructure?

The Hidden Dangers of Vibe Coding and Agentic AI: Why 69% of Organizations Find Vulnerabilities in AI Code

Overview

In April 2026, the ACM warned that AI-assisted coding, especially "vibe coding" using natural language prompts, is rapidly transforming software development but brings serious risks. While it boosts productivity, it often produces insecure code with hidden vulnerabilities and risky third-party dependencies. Manual reviews struggle to keep up, causing governance and accountability to break down as developers lose deep understanding of AI-generated code. Meanwhile, autonomous agentic AI systems amplify threats by enabling dangerous prompt injection attacks and cascading system failures. These issues stem from inherent AI hallucinations that generate flawed code. The ACM calls for rigorous testing, human oversight, and strong policies to ensure AI-driven development is safe and sustainable.

...