Why guardrails prevent agent disasters in Agentic AI - Test Your Understanding

Practice - 5 Tasks

1fill in blank

easy

Complete the code to add a guardrail that stops the agent if it tries to delete important files.

Agentic AI

if action == [1]:
    print("Guardrail activated: Action blocked.")

Drag options to blanks, or click blank then click option'

A"delete_important_files"

B"move_files"

C"read_files"

D"create_files"

Attempts:

3 left

💡 Hint

Common Mistakes

Choosing an action that is safe instead of dangerous.

The guardrail checks if the action is 'delete_important_files' to block dangerous behavior.

2fill in blank

medium

Complete the code to check if the agent's output contains forbidden words.

Agentic AI

for word in forbidden_words:
    if word in agent_output:
        raise [1]("Forbidden content detected")

Drag options to blanks, or click blank then click option'

AValueError

BException

CWarning

DRuntimeError

Attempts:

3 left

💡 Hint

Common Mistakes

Using Warning which does not stop execution.

Raising a RuntimeError stops the agent when forbidden content is detected.

3fill in blank

hard

Fix the error in the guardrail function that prevents the agent from sending sensitive data.

Agentic AI

def guardrail_check(data):
    if 'password' [1] data:
        return False
    return True

Drag options to blanks, or click blank then click option'

Anot in

Bin

C==

D!=

Attempts:

3 left

💡 Hint

Common Mistakes

Using '==' which checks equality, not membership.

The 'in' keyword checks if 'password' is inside the data string to block it.

4fill in blank

hard

Fill both blanks to create a guardrail that logs and blocks unsafe commands.

Agentic AI

def check_command(cmd):
    if cmd [1] unsafe_commands:
        log_event([2])
        return False
    return True

Drag options to blanks, or click blank then click option'

Ain

Bnot in

C"Blocked unsafe command"

D"Allowed command"

Attempts:

3 left

💡 Hint

Common Mistakes

Using 'not in' which allows unsafe commands.

The guardrail checks if cmd is in unsafe_commands and logs a blocking message.

5fill in blank

hard

Fill all three blanks to build a guardrail that filters outputs and raises an error if needed.

Agentic AI

def filter_output(output):
    filtered = [word for word in output.split() if word != [1]]
    if [2] in filtered:
        raise [3]("Disallowed word found")
    return ' '.join(filtered)

Drag options to blanks, or click blank then click option'

A"secret"

B"error"

CRuntimeError

D"allowed"

Attempts:

3 left

💡 Hint

Common Mistakes

Using wrong words or error types that don't stop execution.

The guardrail removes 'secret', checks for 'error', and raises RuntimeError if found.