Skip to content
InfoNews

How organizations can secure their AI-generated code

Published
Record updated
View JSON

Summary

A company's IT manager discovered that a new hire had built a web application handling customer data and private contracts, hosted on Supabase's free plan and deployed from GitHub to Cloudflare, without reviewing the AI-generated code for security. The CEO dismissed the concern. A Carnegie Mellon benchmark called SusVibes found that agents produced working code up to 57% of the time but only 11.8% of solutions were secure.

Mitigation

To treat AI-generated code "as untrusted input that has to earn its way into the codebase," security leaders are moving checks closer to where code is written, running static analysis, secrets scanning and dependency checks on every pull request and increasingly in the developer's editor. Companies are also approving specific AI tools, limiting what assistants can access, and training reviewers to spot common AI mistakes. Specialist human review is applied to code involving access, authentication or sensitive data.