Improper Control of Generation of Code ('Code Injection') (CWE-94)

Improper Control of Generation of Code ('Code Injection')

The product constructs all or part of a code segment using externally-influenced input from an upstream component, but it does not neutralize or incorrectly neutralizes special elements that could modify the syntax or behavior of the intended code segment.

When software allows a user's input to contain code syntax, it might be possible for an attacker to craft the code in such a way that it will alter the intended control flow of the software. Such an alteration could lead to arbitrary code execution.

Prevalence

High

Frequently exploited

Impact

Critical

6 critical-severity rules

Prevention

Documented

10 fix examples

2 Prevention

How to fix this vulnerability

Prevention strategies for Code Injection based on 10 Shoulder detection rules.

🐹 Go View all Go details →

Code Injection via os/exec CRITICAL

Pass user input as template data, never use template.HTML with unsanitized input

+4 -4 go

  package main
  
  import (
      "html/template"
      "net/http"
  )
  
  func handler(w http.ResponseWriter, r *http.Request) {
      userContent := r.FormValue("content")
-     // Vulnerable: user input cast to template.HTML bypasses escaping
-     unsafe := template.HTML("<div>" + userContent + "</div>")
-     tmpl, _ := template.New("page").Parse(`<html>{{.}}</html>`)
-     tmpl.Execute(w, unsafe)
+     // Safe: pass as template data, html/template auto-escapes
+     data := struct{ Content string }{Content: userContent}
+     tmpl, _ := template.New("page").Parse(`<html><div>{{.Content}}</div></html>`)
+     tmpl.Execute(w, data)
  }

LLM Insecure Output Handling HIGH

Validate and sanitize LLM outputs before using in dangerous operations like exec or SQL

+6 -1 go

  resp, _ := client.CreateChatCompletion(ctx, openai.ChatCompletionRequest{
      Messages: []openai.ChatCompletionMessage{{Content: "Generate command for: " + task}},
  })
  cmd := resp.Choices[0].Message.Content
- exec.Command("bash", "-c", cmd).Run()
+ 
+ validCommands := map[string]bool{"ls": true, "pwd": true, "date": true}
+ if !validCommands[cmd] {
+     return fmt.Errorf("invalid command: %s", cmd)
+ }
+ exec.Command(cmd).Run()

Server-Side Template Injection CRITICAL

Use predefined templates and pass user input as template data, never as template code

+9 -5 go

  package main
  
  import (
      "html/template"
      "net/http"
  )
  
- func handler(w http.ResponseWriter, r *http.Request) {
-     tmplStr := r.FormValue("template")
-     // Vulnerable: user input parsed as template code
-     tmpl, _ := template.New("page").Parse(tmplStr)
-     tmpl.Execute(w, nil)
+ // Safe: template is predefined, not from user input
+ var pageTmpl = template.Must(template.ParseFiles("templates/page.html"))
+ 
+ func handler(w http.ResponseWriter, r *http.Request) {
+     name := r.FormValue("name")
+     // Safe: user input passed as data, not template code
+     pageTmpl.Execute(w, map[string]string{
+         "name": name,
+     })
  }

🟨 JavaScript View all JavaScript details →

Code Injection via eval() and Function constructor CRITICAL

Replace eval/Function constructor with safe alternatives like JSON.parse or predefined function maps

+11 -4 javascript

  const express = require('express');
  const app = express();
  
- app.post('/calculate', (req, res) => {
-   const expression = req.body.expression;
-   const result = eval(expression);
-   res.json({ result });
+ const operations = {
+   add: (a, b) => a + b,
+   subtract: (a, b) => a - b,
+   multiply: (a, b) => a * b,
+ };
+ 
+ app.post('/calculate', (req, res) => {
+   const { op, a, b } = req.body;
+   const fn = operations[op];
+   if (!fn) return res.status(400).json({ error: 'Invalid operation' });
+   res.json({ result: fn(Number(a), Number(b)) });
  });

TypeScript Unsafe Decorator Usage HIGH

Use static values for decorator parameters and avoid eval(), global modifications, or user input in decorators

+16 -14 javascript

- function DynamicRole(roleExpression: string) {
-   return function (target: any, key: string, desc: PropertyDescriptor) {
-     const original = desc.value;
-     desc.value = function (...args: any[]) {
-       if (eval(roleExpression)) {
-         return original.apply(this, args);
-       }
-       throw new Error('Unauthorized');
-     };
-   };
- }
- 
- class AdminController {
-   @DynamicRole("user.role === 'admin'")
+ enum Role { Admin = 'admin', User = 'user' }
+ 
+ function RequireRole(...roles: Role[]) {
+   return function (target: any, key: string, desc: PropertyDescriptor) {
+     const original = desc.value;
+     desc.value = function (...args: any[]) {
+       if (!roles.includes(this.currentUser?.role)) {
+         throw new Error('Unauthorized');
+       }
+       return original.apply(this, args);
+     };
+   };
+ }
+ 
+ class AdminController {
+   @RequireRole(Role.Admin)
    deleteUser() { /* ... */ }
  }

🐍 Python View all Python details →

Code Injection via eval/exec CRITICAL

Use ast.literal_eval() for safe evaluation or avoid eval/exec entirely

+10 -6 python

- from flask import request
- 
- @app.route('/calc')
- def calculate():
-     expression = request.args.get('expr')
-     result = eval(expression)
+ import ast
+ from flask import request, abort
+ 
+ @app.route('/calc')
+ def calculate():
+     expression = request.args.get('expr', '')
+     try:
+         result = ast.literal_eval(expression)
+     except (ValueError, SyntaxError):
+         abort(400, 'Invalid expression')
      return str(result)

Dangerous Function Usage CRITICAL

Replace eval/exec with ast.literal_eval, JSON parsing, or subprocess with shell=False

+7 -6 python

- from flask import request
- 
- @app.route('/calculate')
- def calculate():
-     expr = request.args.get('expr')
-     result = eval(expr)
+ import ast
+ from flask import request
+ 
+ @app.route('/calculate')
+ def calculate():
+     expr = request.args.get('expr')
+     result = ast.literal_eval(expr)
      return {'result': result}

LLM Insecure Output Handling HIGH

Validate and sanitize LLM outputs with Pydantic before using in dangerous operations like eval, exec, or SQL

+18 -6 python

- response = openai.chat.completions.create(
-     model='gpt-4',
-     messages=[{'role': 'user', 'content': user_request}]
- )
- generated_code = response.choices[0].message.content
- result = eval(generated_code)
+ from pydantic import BaseModel, validator
+ import re
+ 
+ class ValidatedOutput(BaseModel):
+     expression: str
+ 
+     @validator('expression')
+     def validate_expression(cls, v):
+         if not re.fullmatch(r'[a-zA-Z0-9\s\+\-\*\/\(\)\.]+', v):
+             raise ValueError('Invalid expression format')
+         return v
+ 
+ response = openai.chat.completions.create(
+     model='gpt-4',
+     messages=[{'role': 'user', 'content': user_request}]
+ )
+ validated = ValidatedOutput(expression=response.choices[0].message.content)
+ result = process_validated_expression(validated.expression)

Key Practices

treated as untrusted input since: - Prompt injection attacks can manipulate AI responses - LLMs can hallucinate and produce unexpected outputs - Model behavior may change between versions Dangerous operations include: - Code execution (eval, Function, vm
avoided or heavily restricted
treated as untrusted input since: - Prompt injection attacks can manipulate AI responses - LLMs can hallucinate and produce unexpected outputs - Model behavior may change between versions Dangerous operations include: - Code execution (eval, exec, compile) - Command execution (os

3 Detection

Find vulnerabilities in your code

Use Shoulder to scan your codebase for Improper Control of Generation of Code ('Code Injection') patterns. 10 rules.

terminal

# Scan with Shoulder CLI
npx @shoulderdev/cli trust --cwe=94

# Or scan entire project
npx @shoulderdev/cli trust .

Detection Rules (10)

🐍 Python 4 rules

Code Injection via eval/exec CRITICAL

Detects untrusted user input flowing into code evaluation functions (eval, exec, compile).

Dangerous Function Usage CRITICAL

Detects usage of dangerous Python functions that can lead to arbitrary code execution: eval(), exec(), compile(), __import__() with user input, or pickle deserialization. These should be avoided or heavily restricted.

LLM Insecure Output Handling HIGH

Detects LLM/AI outputs being used directly in dangerous operations without proper validation or sanitization. OWASP LLM02 - Insecure Output Handling. LLM outputs should be treated as untrusted input since: - Prompt injection attacks can manipulate AI responses - LLMs can hallucinate and produce unexpected outputs - Model behavior may change between versions Dangerous operations include: - Code execution (eval, exec, compile) - Command execution (os.system, subprocess) - SQL queries (cursor.execute, raw queries) - Template rendering (Jinja2, Django templates) - File operations (open, write, unlink) - Deserialization (pickle, yaml.load)

Server-Side Template Injection (SSTI) CRITICAL

Detects user input used directly in template rendering, allowing arbitrary code execution.

🐹 Go 3 rules

Code Injection via os/exec CRITICAL

Detects user input flowing to template functions that bypass HTML escaping.

LLM Insecure Output Handling HIGH

Detects LLM outputs used directly in dangerous operations like command execution or SQL queries without validation.

Server-Side Template Injection CRITICAL

User input passed directly to template.Parse without sanitization.

🔷 Typescript 3 rules

Code Injection via eval() and Function constructor CRITICAL

Detects user input flowing to code execution functions like eval() or Function constructor.

LLM Insecure Output Handling HIGH

Detects LLM/AI outputs being used directly in dangerous operations without proper validation or sanitization. OWASP LLM02 - Insecure Output Handling. LLM outputs should be treated as untrusted input since: - Prompt injection attacks can manipulate AI responses - LLMs can hallucinate and produce unexpected outputs - Model behavior may change between versions Dangerous operations include: - Code execution (eval, Function, vm.runInContext) - Command execution (exec, spawn, execSync) - SQL queries (database operations) - HTML rendering (innerHTML, document.write) - File operations (writeFile, unlink) - Network requests (fetch, axios with LLM-generated URLs)

TypeScript Unsafe Decorator Usage HIGH

Decorators that use eval(), modify global state, or accept user input as parameters enable code injection, prototype pollution, and authorization bypass.

🟨 Javascript 2 rules

Code Injection via eval() and Function constructor CRITICAL

Detects user input flowing to code execution functions like eval() or Function constructor.

LLM Insecure Output Handling HIGH

Detects LLM/AI outputs being used directly in dangerous operations without proper validation or sanitization. OWASP LLM02 - Insecure Output Handling. LLM outputs should be treated as untrusted input since: - Prompt injection attacks can manipulate AI responses - LLMs can hallucinate and produce unexpected outputs - Model behavior may change between versions Dangerous operations include: - Code execution (eval, Function, vm.runInContext) - Command execution (exec, spawn, execSync) - SQL queries (database operations) - HTML rendering (innerHTML, document.write) - File operations (writeFile, unlink) - Network requests (fetch, axios with LLM-generated URLs)

4 Warning Signs

What to watch for in code reviews

These patterns indicate potential Improper Control of Generation of Code ('Code Injection') vulnerabilities. Look for these during code reviews and security audits.

🟠

LLM output flows to ... without validation go-llm-insecure-output-handling

🟠

LLM outputs used directly in dangerous operations like command execution or SQL queries without vali go-llm-insecure-output-handling

🟠

LLM/AI outputs being used directly in dangerous operations without proper validation or sanitization javascript-llm-insecure-output-handling

🟠

Decorator '...' executes unsafe code or accesses global state. This can lead to code injection or unauthorized access. typescript-unsafe-decorator

🔴

user input flowing to template functions that bypass HTML escaping go-code-injection

🔴

user input flowing to code execution functions like eval() or Function constructor javascript-code-injection

🔴

untrusted user input flowing into code evaluation functions (eval, exec, compile) python-code-injection

🔴

usage of dangerous Python functions that can lead to arbitrary code execution: eval(), exec(), compi python-dangerous-functions

🔍

Scan your codebase for Improper Control of Generation of Code ('Code Injection')

Shoulder CLI finds vulnerable patterns across your entire codebase.

npx shoulder trust Browse all rules