Security & Compliance for AI-Assisted Coding

AI will happily generate insecure or non-compliant code if you let it.
Treat AI like a smart but untrusted junior developer — assume it will make these mistakes unless you prevent and detect them.

1. What AI Gets Wrong

These are the security mistakes AI tools make over and over again:

AI habit	Example
Concatenates user input into queries	`f"SELECT * FROM users WHERE name = '{name}'"`
Hard-codes secrets	`API_KEY = "sk-proj-abc123..."`
Skips auth on new endpoints	Generates a route with no middleware
Logs PII	`logger.info(f"User {email} did {action}")`
Suggests outdated or vulnerable packages	Recommends a library that was abandoned 2 years ago
Invents APIs that don’t exist	Calls a method that looks right but was never part of the library

Warning

AI-generated code often looks secure at first glance. The patterns are idiomatic, the naming is clean — but the security holes are real. Always review with security in mind.

2. What to Watch For (AI-Specific)

AI adds dependencies you didn’t ask for

AI frequently suggests packages it “remembers” from training data. These may be outdated, unmaintained, or even typosquatted.

Before accepting a new dependency:

Does it actually exist? Check npm/PyPI/etc.
When was it last updated?
How many downloads / maintainers does it have?
Is there a known vulnerability? (Check GitHub Advisories)

AI doesn’t know your auth setup

AI generates endpoints and forgets your middleware, your role model, or your session handling. Every new route or API it creates is a potential open door.

Rule: If AI generates a new endpoint, the first question is: “Where is the auth check?”

AI puts secrets in code

AI has seen thousands of tutorials with hard-coded keys. It will happily reproduce that pattern. It will also suggest committing .env files or logging tokens for “debugging.”

Rule: Never accept AI code that contains string literals looking like keys, tokens, or passwords.

AI over-collects data

When asked to “log user actions” or “track usage,” AI tends to log everything — emails, IPs, full request bodies. This is a GDPR problem waiting to happen.

Rule: Log event types and anonymized IDs. Never log PII unless you have a documented legal basis.

3. Catching AI Mistakes with GitHub

GitHub has built-in tools that are especially useful when AI writes your code, because they catch exactly the mistakes AI tends to make.

Secret Scanning + Push Protection

AI loves to hard-code secrets. GitHub secret scanning catches them.

Settings → Security → Code security and analysis → Secret scanning

Scans for known secret patterns (AWS keys, GitHub tokens, etc.)
Push protection blocks commits containing secrets before they reach the repo

Tip

Enable push protection. It’s easier to block a secret than to rotate one that’s already in git history.

Dependabot

AI suggests packages it saw in training data — which may have known vulnerabilities by now.

Settings → Security → Code security and analysis → Dependabot

Dependabot Alerts — flags known vulnerabilities in your dependencies
Security Updates — auto-opens PRs to fix vulnerable packages

# .github/dependabot.yml
version: 2
updates:
  - package-ecosystem: "npm"
    directory: "/"
    schedule:
      interval: "weekly"

Code Scanning (CodeQL)

AI generates injection vulnerabilities, XSS, path traversal — CodeQL catches these statically.

# .github/workflows/codeql.yml
name: "CodeQL"
on:
  push:
    branches: [main]
  pull_request:
    branches: [main]

jobs:
  analyze:
    runs-on: ubuntu-latest
    permissions:
      security-events: write
    strategy:
      matrix:
        language: [javascript, python]
    steps:
      - uses: actions/checkout@v4
      - uses: github/codeql-action/init@v3
        with:
          languages: ${{ matrix.language }}
      - uses: github/codeql-action/analyze@v3

Branch Protection

AI generates fast. That’s dangerous without a review gate.

Settings → Branches → Add branch protection rule

Require PR reviews before merging
Require status checks to pass (CI + CodeQL)
Do not allow bypassing these settings

Why this matters for AI: The speed of AI-generated code makes it tempting to skip review. Branch protection ensures nothing lands on main without human eyes and green checks.

4. What Goes Into AI Prompts (Compliance)

When using third-party AI tools (Copilot, Cursor, ChatGPT, Claude), your prompts are data you’re sending to an external service.

Never put in prompts	Why
Production secrets, API keys	They may be stored or logged by the provider
Customer PII (names, emails, IDs)	GDPR/DSGVO violation — you have no DPA for “pasting into ChatGPT”
Proprietary source code	May be used for model training (check your plan’s terms)
Internal architecture docs	Competitive risk

For teams:

Define which AI tools are approved and on which plans (enterprise vs. free)
Prefer enterprise offerings that guarantee no training on your code
Require Data Processing Agreements (DPAs) with AI vendors
Consider on-prem or VPC-hosted models for sensitive environments

5. Teach Your AI the Rules

Encode security rules in your project’s AI instruction files so the AI follows them by default:

Files: AGENTS.md, .cursorrules, copilot-instructions.md, CLAUDE.md

## Security Rules
- Never log PII or secrets
- Always use parameterized queries — no string concatenation for SQL
- All new endpoints require authentication middleware
- Do not hard-code API keys, tokens, or passwords
- Do not add dependencies without checking for known vulnerabilities

This won’t make AI perfect, but it significantly reduces how often you have to correct the same mistakes.

Checklist

Before merging AI-assisted changes:

No hard-coded secrets, tokens, or credentials
New endpoints have authentication and authorization checks
Database access uses parameterized queries
Logs contain no PII or secrets
New dependencies are verified (exist, maintained, no known vulnerabilities)
GitHub secret scanning + push protection enabled
CodeQL / SAST checks pass
At least one human has reviewed the change with security in mind
No customer data or secrets were pasted into AI prompts