# Improper Access Control (CWE-284)

The product does not restrict or incorrectly restricts access to a resource from an unauthorized actor.

- Prevalence: High Frequently exploited
- Impact: High 3 high-severity rules
- Prevention: Documented 4 fix examples

**OWASP:** Broken Access Control (A01:2021-Broken Access Control) - #1

## Description

Access control involves determining which subjects can access which objects. When access control is implemented incorrectly, it can lead to unauthorized access to sensitive data or functionality.

## Prevention

Prevention strategies for Improper Access Control based on 4 Shoulder detection rules.

### Go

Validate tool inputs against strict schemas and use an allowlist for permitted tools

### Node.js

Validate tool inputs against schemas and use allowlists for permitted tools

### Kubernetes

Define NetworkPolicy resources to restrict pod-to-pod traffic and enforce network segmentation

### Python

Use Pydantic for tool input validation and maintain a strict allowlist for permitted tools

## Warning Signs

- [HIGH] Insecure plugin implementation: ...
- [HIGH] insecure plugin/function calling implementations in AI/LLM systems without proper validation
- [MEDIUM] Workload has no NetworkPolicy for network segmentation
- [MEDIUM] Kubernetes deployments without associated NetworkPolicy resources

## Consequences

- Read Application Data
- Modify Application Data
- Execute Unauthorized Code
- Gain Privileges

## Mitigations

- Implement proper access control checks on all resources
- Use the principle of least privilege
- Enforce access controls server-side, not just in the UI

## Detection

- Total rules: 4
- Languages: go, javascript, typescript, kubernetes, yaml, python

## Rules by Language

### Go (1 rules)

- **LLM Insecure Plugin Design** [HIGH]: Detects insecure plugin/function calling implementations in AI/LLM systems without proper validation.
  - Remediation: Validate tool inputs against strict schemas and use an allowlist for permitted tools.

```go
if _, ok := toolRegistry[toolCall.Name]; !ok {
    return errors.New("unknown tool")
}
```

Learn more: https://shoulder.dev/learn/go/cwe-284/llm-insecure-plugin

### Javascript (1 rules)

- **LLM Insecure Plugin Design** [HIGH]: Detects insecure plugin/function calling implementations in AI/LLM systems.
OWASP LLM07 - Insecure Plugin Design.

Insecure plugin design can lead to:
- Remote code execution via tool/function calls
- Unauthorized data access through plugins
- Privilege escalation via overly permissive tools
- SSRF through URL-handling plugins
- Command injection through shell plugins

This rule detects:
- Function calling without input validation
- Dynamic function execution from LLM output
- Plugin execution without access control
- Dangerous functions exposed to LLM
  - Remediation: Validate tool inputs against schemas and use allowlists for permitted tools.

```javascript
if (!allowedTools.includes(name)) throw new Error('Unknown tool');
const validate = ajv.compile(toolSchemas[name]);
if (!validate(JSON.parse(args))) throw new Error('Invalid arguments');
```

Learn more: https://shoulder.dev/learn/javascript/cwe-284/llm-insecure-plugin

### Typescript (1 rules)

- **LLM Insecure Plugin Design** [HIGH]: Detects insecure plugin/function calling implementations in AI/LLM systems.
OWASP LLM07 - Insecure Plugin Design.

Insecure plugin design can lead to:
- Remote code execution via tool/function calls
- Unauthorized data access through plugins
- Privilege escalation via overly permissive tools
- SSRF through URL-handling plugins
- Command injection through shell plugins

This rule detects:
- Function calling without input validation
- Dynamic function execution from LLM output
- Plugin execution without access control
- Dangerous functions exposed to LLM
  - Remediation: Validate tool inputs against schemas and use allowlists for permitted tools.

```javascript
if (!allowedTools.includes(name)) throw new Error('Unknown tool');
const validate = ajv.compile(toolSchemas[name]);
if (!validate(JSON.parse(args))) throw new Error('Invalid arguments');
```

Learn more: https://shoulder.dev/learn/javascript/cwe-284/llm-insecure-plugin

### Kubernetes (1 rules)

- **Missing Network Policy** [MEDIUM]: Detects Kubernetes deployments without associated NetworkPolicy resources.
  - Remediation: Define a NetworkPolicy to control pod network access.

```yaml
kind: NetworkPolicy
spec:
  podSelector: {}
  policyTypes: [Ingress]
```

Learn more: https://shoulder.dev/learn/kubernetes/cwe-284/missing-network-policy

### Yaml (1 rules)

- **Missing Network Policy** [MEDIUM]: Detects Kubernetes deployments without associated NetworkPolicy resources.
  - Remediation: Define a NetworkPolicy to control pod network access.

```yaml
kind: NetworkPolicy
spec:
  podSelector: {}
  policyTypes: [Ingress]
```

Learn more: https://shoulder.dev/learn/kubernetes/cwe-284/missing-network-policy

### Python (1 rules)

- **LLM Insecure Plugin Design** [HIGH]: Detects insecure plugin/function calling implementations in AI/LLM systems.
OWASP LLM07 - Insecure Plugin Design.

Insecure plugin design can lead to:
- Remote code execution via tool/function calls
- Unauthorized data access through plugins
- Privilege escalation via overly permissive tools
- SSRF through URL-handling plugins
- Command injection through shell plugins
  - Remediation: Use Pydantic for tool input validation and maintain an allowlist of tools.

```python
from pydantic import BaseModel, Field

class SearchArgs(BaseModel):
    query: str = Field(max_length=100, pattern=r'^[a-zA-Z0-9\s]+$')

ALLOWED_TOOLS = {'search_products', 'get_weather'}

def execute(tool_call):
    if tool_call.function.name not in ALLOWED_TOOLS:
        raise ValueError('Unknown tool')
    args = SearchArgs.parse_raw(tool_call.function.arguments)
    return handlers[tool_call.function.name](args)
```

Learn more: https://shoulder.dev/learn/python/cwe-284/llm-insecure-plugin