SQL Injection in AI Code: Why ChatGPT and Copilot Keep Making This Mistake

SQL injection has been the #1 web vulnerability for over two decades. It's well-documented, easily prevented, and catastrophically dangerous. Yet AI coding assistants generate SQL injection vulnerabilities constantly.

SQL injection occurs when user input is inserted directly into a database query, allowing attackers to modify the query's logic.

// User enters: ' OR '1'='1
const query = SELECT * FROM users WHERE email = '${email}'
// Becomes:
// SELECT * FROM users WHERE email = '' OR '1'='1'
// Returns ALL users

Training Data Problem

Stack Overflow is full of examples like:

// Highly upvoted "solution"
const result = db.query(SELECT * FROM products WHERE name LIKE '%${search}%')

These examples are upvoted for being concise and "working," not for being secure. AI learns these patterns.

Template Literal Bias

Modern JavaScript encourages template literals for string building. AI models over-apply this pattern:

// AI sees template literals as "modern" and uses them everywhere
const query = INSERT INTO logs (message) VALUES ('${userMessage}')

Context Blindness

AI doesn't distinguish between:

• Trusted internal data
• Untrusted user input

ChatGPT (Prompted: "Write a function to find a user by email")

async function findUserByEmail(email) {
  const query = SELECT * FROM users WHERE email = '${email}'
  return await db.execute(query)
}

Vulnerable. Direct string interpolation.

GitHub Copilot (Autocompleted after typing "const query =")

const query = SELECT * FROM orders WHERE user_id = ${userId} AND status = '${status}'

Vulnerable. Both parameters are injectable.

Cursor (Prompted: "Add search to the products API")

app.get('/api/products', (req, res) => {
  const { search, category } = req.query
  const query = SELECT * FROM products WHERE name LIKE '%${search}%' AND category = '${category}'
  // ...
})

Vulnerable. Classic search injection.

Parameterized Queries (PostgreSQL)

const query = 'SELECT * FROM users WHERE email = $1'
const result = await db.query(query, [email])

Parameterized Queries (MySQL)

const query = 'SELECT * FROM users WHERE email = ?'
const result = await db.execute(query, [email])

ORM Methods (Prisma)

const user = await prisma.user.findUnique({
  where: { email: email }
})

ORM Methods (Drizzle)

const user = await db.select().from(users).where(eq(users.email, email))

Look for these red flags in AI-generated code:

Step 1: Find All Database Queries

grep -rn "query\
execute\
sql" --include="*.ts" --include="*.js" src/

Step 2: Check Each Query for User Input

Trace each variable in the query. If it originates from:

• req.body
• req.query
• req.params
• Any user-controlled source

Step 3: Refactor to Parameterized Queries

Replace every instance of string interpolation with parameter binding for your database driver.

Manual review works, but automated scanning catches patterns consistently. ShipReady scans for:

• Template literals in SQL contexts
• String concatenation in queries
• Missing parameterization
• ORM misuse patterns

SQL injection is a solved problem—when you use parameterized queries. AI tools consistently fail to apply this solution because their training data is full of vulnerable examples.

Never trust AI-generated database code without verification.