Django raw SQL: safe parameters versus string formatting

Django QuerySets are safe from SQL injection because the database driver escapes the parameters. Raw SQL is where it goes wrong. The rule is simple: pass user values as separate parameters, never build the SQL string with formatting or f-strings.

What is safe by default

According to the Django security docs, queries built through the ORM are parameterized. You write Person.objects.filter(last_name=user_input) and the driver handles escaping. You do not need to think about quoting.

Where the risk starts

The docs name the escape hatches: raw(), cursor.execute(), extra() and RawSQL. Use them sparingly, and always pass user-controlled values as parameters.

# Safe: parameters passed separately
Person.objects.raw("SELECT * FROM app_person WHERE last_name = %s", [user_input])

# Unsafe: string interpolation
Person.objects.raw(f"SELECT * FROM app_person WHERE last_name = '{user_input}'")

Notice that the safe form uses %s as a placeholder, not Python string formatting. The list at the end is what carries the value to the driver.

Why formatting is dangerous even when it looks harmless

An f-string pastes the value into the SQL text before the database parses it. A value containing a quote can end your string literal and add its own SQL. A parameter, by contrast, is sent as data and never parsed as SQL.

Placeholders cannot do everything

Placeholders work for values. They do not work for table names, column names or sort direction. If a user picks a sort column, map the input to a fixed allowlist in your code and put only your own constant into the SQL.

ALLOWED_SORT = {"name": "last_name", "created": "created_at"}
column = ALLOWED_SORT.get(request.GET.get("sort"), "created_at")

A review habit that catches most cases

Search the codebase for raw(, .extra(, RawSQL and cursor.execute. For each hit, ask two questions: does any value come from a request, and is it passed as a parameter. If a value is inside the SQL string, fix it.

Dependency scanning is a separate layer. A flaw in a library you did not write cannot be found by reading your own raw queries, and Safeguard's SBOM and dependency scanning covers that side. The free scan is at gold.safeguard.sh/scan.

Quick check

Sources

• Django security overview

Last reviewed: 19 October 2026.