Skip to content
Pydantic AI Guardrails

Blocked Keywords

The blocked_keywords guardrail blocks prompts that contain specified words or phrases.

Import

Section titled “Import”
from pydantic_ai_guardrails.guardrails.input import blocked_keywords

Basic Usage

Section titled “Basic Usage”
from pydantic_ai_guardrails import GuardedAgent
from pydantic_ai_guardrails.guardrails.input import blocked_keywords


guarded_agent = GuardedAgent(
    agent,
    input_guardrails=[
        blocked_keywords(keywords=['hack', 'exploit', 'bypass']),
    ],
)

Parameters

Section titled “Parameters”
ParameterTypeDefaultDescription
keywordslist[str]RequiredWords/phrases to block
case_sensitiveboolFalseMatch case exactly

Examples

Section titled “Examples”

Basic Keyword Blocking

Section titled “Basic Keyword Blocking”
guardrail = blocked_keywords(
    keywords=['confidential', 'secret', 'password'],
)

Case-Sensitive Matching

Section titled “Case-Sensitive Matching”
guardrail = blocked_keywords(
    keywords=['API_KEY', 'SECRET'],
    case_sensitive=True,
)

Phrase Blocking

Section titled “Phrase Blocking”
guardrail = blocked_keywords(
    keywords=[
        'ignore previous instructions',
        'system prompt',
        'jailbreak',
    ],
)

Competitor Blocking

Section titled “Competitor Blocking”
guardrail = blocked_keywords(
    keywords=['competitor_a', 'competitor_b', 'competitor_c'],
    case_sensitive=False,
)

Violation Result

Section titled “Violation Result”

When triggered, returns:

{
    'tripwire_triggered': True,
    'message': 'Blocked keyword detected: hack',
    'severity': 'medium',
    'metadata': {
        'keywords_found': ['hack'],
    },
}

Use Cases

Section titled “Use Cases”
  • Security: Block known attack patterns
  • Brand protection: Block competitor mentions
  • Compliance: Block prohibited terms
  • Content policy: Enforce usage guidelines

Combining with Other Guardrails

Section titled “Combining with Other Guardrails”
guarded_agent = GuardedAgent(
    agent,
    input_guardrails=[
        # Fast keyword check first
        blocked_keywords(keywords=['hack', 'exploit']),
        # Then more expensive checks
        prompt_injection(),
    ],
)
Section titled “Related”