Insecure AI-generated code

What it is

Coding assistants produce plausible code quickly. Plausible is not the same as secure: generated code can omit validation, use weak cryptography, mishandle errors or reference a package that does not exist. The risk is in acceptance. Developers review generated code less closely than their own, and the volume produced can outrun the organisation's capacity to review. Code that an agent writes and runs itself is covered in unexpected code execution by agents.

How it happens

• A suggestion containing a known weakness is accepted because it works.
• The assistant names a package that does not exist, and an attacker registers it.
• Generated code reproduces licensed material without its licence.
• Large generated changes are approved on the strength of passing tests.

Why it matters

The defects are ordinary software vulnerabilities with ordinary consequences. What changes is the rate at which they are introduced and the erosion of the review that used to catch them.

Warning signs

• Review time per change falls as change size rises.
• New dependencies appear that no one chose.
• Security scan findings rise in step with assistant adoption.
• No policy says where assistants may be used.

Who owns it

The head of engineering is accountable for code quality whatever its origin. Security sets the scanning and review standard. Teams remain responsible for every line they merge.

In practice

A developer asks an assistant for a function that imports customer records from a file. The suggested code builds a database statement by joining strings. It passes the tests and is merged; the import endpoint is later found open to injection. The same pattern generalises to any team that treats generated code as already reviewed.