Database Security

Rules (for AI agents)

ALWAYS

• Bind every untrusted data value through the driver or ORM parameter API. Never place it into SQL with language-level interpolation or concatenation — +, f-strings, % formatting, .format(), JavaScript template interpolation. Driver placeholders (%s, ?, $1, :name, @name) are the binding mechanism, not string formatting; confusing the two is how this rule is most often misread. The same applies to values read back out of the database and used to build a later query: stored input is still input, and second-order injection is this bug with a delay.
• Prefer the ORM's structured query-expression API over textual SQL. Where raw SQL is genuinely needed, use that framework's documented parameter-binding mechanism and never interpolate untrusted values into it. references/orm-safe-query-examples.md carries the verified form per framework.
• For SQL structure that cannot be a bind parameter — table and column identifiers, sort direction, sometimes the operator itself — map the external value to a fixed operator-controlled allowlist. Never copy an untrusted identifier or SQL fragment into the statement, escaped or not.
• Treat a NoSQL query as a structure, not a string. Build filters with the driver's typed query objects, and coerce every request-supplied value to its expected scalar type before it reaches a filter. A JSON body substituting {"$ne": null} where a string was expected turns an equality check into a match-all — operator injection needs no quote to break out of.
• Give each application its own database identity with the minimum grants it needs, and separate runtime privileges from migration privileges. The runtime identity must not receive schema-owner or DDL rights merely because migrations require them.
• For shared-table multi-tenancy, enforce tenant isolation at a deliberate security boundary. Where database RLS is used: confirm the application role cannot bypass it — a superuser, a BYPASSRLS role, and the table owner itself all bypass policies unless the table is set to FORCE ROW LEVEL SECURITY — test write policies as well as read, and bind the tenant context transaction-locally. A session-scoped tenant context that survives a connection's return to the pool serves tenant A's rows to tenant B.
• Authenticate the database server, do not merely encrypt to it. Encryption without certificate verification (sslmode=require, MySQL sslMode=REQUIRED, TrustServerCertificate=True) stops passive capture but not a redirected connection. Where the client is what establishes trust, use a verifying mode — PostgreSQL verify-full, MySQL Connector/J VERIFY_IDENTITY, Microsoft.Data.SqlClient with TrustServerCertificate=False — against the deployment's trusted CA configuration.
• Obtain database credentials from an operator-controlled secret-management mechanism; prefer workload identity or short-lived dynamic credentials where the platform supports them. A runtime-injected environment variable is an acceptable fallback, not the target — it is readable through process inspection, crash dumps, and many orchestrator APIs. Any static credential must support rotation and revocation. See secret-detection for what counts as a committed credential.

NEVER

• Manually quote or escape an untrusted SQL value as a substitute for the framework's documented parameterization mechanism.
• Use an ORM raw-query method (.raw(), .objects.raw(), $queryRawUnsafe, session.execute(text(...))) with an interpolated f-string or template literal. Raw SQL is not the problem; interpolation into it is.
• Evaluate server-side JavaScript over untrusted input ($where, mapReduce, $accumulator). There is no safe form of it — it is code execution inside the database.
• Run application workloads as the database superuser: root, postgres, sa.
• Grant destructive DDL capability to the normal application runtime identity.
• Send database traffic across an untrusted network without authenticated encryption, or disable certificate verification to work around a certificate error.
• Expose a database listener broadly to the public internet. Restrict reachability to explicitly authorized application and administrative sources, and prefer private connectivity for application traffic where the platform supports it.
• Emit database credentials, connection strings containing secrets, or sensitive bound values to logs. logging-security owns the shape this should take — statement template, parameter names, a hashed value identifier — and the redactor that enforces it.

KNOWN FALSE POSITIVES

• Encryption without client-side certificate verification is not a finding where the channel is already authenticated by other means: a Unix domain socket (where sslmode does not apply at all), a cloud SQL proxy or sidecar terminating an authenticated tunnel, or a service mesh enforcing mTLS. The finding is an unauthenticated connection across a network, not the literal string sslmode=require.
• Some ORMs and DB-API drivers (Django, SQLAlchemy 1.x, psycopg2) use %s as a parameter marker, not a Python format placeholder. cursor.execute(sql, [value]) with an unquoted %s is the safe form, not a violation.
• Trusted analyst-authored ad-hoc SQL in a dedicated analytics environment may intentionally contain dynamic SQL structure. Keep that capability behind explicit authorization and appropriately scoped data access. A read-only account limits modification but does not make untrusted query fragments safe, and does not prevent unauthorized reads.
• Static SQL with no untrusted dynamic values — SELECT 1, fixed schema introspection, constant migration statements — does not need artificial bind parameters to satisfy this skill.

Context (for humans)

Injection has remained a recurring OWASP Top 10 risk across editions. SQL injection stays a high-impact member of that category even as the category's relative ranking moves between releases, and it has not gone away because the failure mode is easy by default: any language with string concatenation lets you produce a query. AI assistants readily generate "works in dev" code that interpolates user input — particularly for sort columns, dynamic filters, and pagination, which are exactly the places a bind parameter cannot be used and an allowlist has to be.

The NoSQL variant is worth naming separately because it does not look like injection. Nothing is quoted, nothing is escaped, and there is no string to break out of — the attacker replaces a scalar with an operator object and the query's shape changes. Reviewers who have learned to spot concatenated SQL will read straight past it.

References

• references/verifying-findings.md — confirm or refute a finding, then lock it
• references/orm-safe-query-examples.md — the verified safe call per framework, placeholder styles, and the allowlist pattern for identifiers
• rules/sql_injection_sinks.json
• rules/orm_safe_patterns.json
• OWASP SQL Injection Prevention Cheat Sheet.
• OWASP NoSQL Security Cheat Sheet.
• CWE-89 — SQL Injection.
• CWE-943 — Data query logic injection.
• PostgreSQL Row-Level Security.