CCDV-F Security Practice Test 2 — 15 Questions

Question 1

A healthcare company uses the Anthropic API to summarize patient messages. The security team requires that protected health information never leaves the company's controlled network boundary and that Anthropic cannot retain any prompt or completion. Which deployment approach best satisfies both requirements?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Deploy Claude through Amazon Bedrock in the company's AWS account and configure a VPC endpoint so traffic stays on the private network.. Meeting both a strict network boundary and a no-retention requirement with Claude means running the model inside an environment the company controls at the network layer. Amazon Bedrock with an AWS PrivateLink VPC endpoint keeps API calls on private AWS networking and provides contractual data-handling terms that exclude training and storage of customer prompts and outputs.

Answer

Use the public Anthropic API with a zero-data-retention agreement and rely on TLS 1.3 in transit.

Answer

Self-host an open-weights model on-premises and route all summarization through it instead of Claude.

Answer

Use the Anthropic API with client-side field-level encryption of PHI fields before sending and decrypt only in the client.

Question 2

Which of the following is the most secure way to handle API keys in an Anthropic-integrated cloud application?

Accepted Answer

Use a cloud-native secret management service to inject keys at runtime.. Managing secrets securely is a standard security practice. Using a dedicated secret manager allows for automatic rotation, granular access control, and audit logging. This prevents the keys from being stored in version control or plain text files, reducing the risk of unauthorized access. It is the industry-standard approach for protecting credentials used by distributed cloud applications.

Answer

Store keys in an encrypted JSON file within the application directory.

Answer

Define keys as environment variables in the Dockerfile during build.

Answer

Hardcode the keys as constants in the configuration file.

Question 3

Refer to the exhibit. An application suddenly begins receiving this error in production. What is the most immediate security-focused action to take?

Accepted Answer

Immediately revoke the current API key and generate a new one.. An authentication error indicates that the currently used key is either revoked, expired, or invalid. In a production environment, this is a major red flag that could signal a credential compromise. The most responsible action is to treat the key as compromised, revoke it immediately in the console, and rotate to a new key to protect the integrity of the application's API interactions.

Answer

Retry the request with an exponential backoff strategy.

Answer

Check if the API billing limit has been reached.

Answer

Hardcode the master account key to restore service quickly.

Question 4

An organization wants to ensure that Claude's responses do not contain harmful or inappropriate content. What is the recommended strategy for output control?

Accepted Answer

Implement a post-processing step to validate output against safety guidelines.. Relying on model training alone is not a complete solution. A layered approach involves using a system prompt to define the tone and safety boundaries, followed by a post-processing filter that checks the model's output against a list of blocked terms or sentiment guidelines. This combination ensures that the model remains within the desired persona while maintaining an external safety check for high-risk applications.

Answer

Only rely on Anthropic's built-in safety filters.

Answer

Ask the user to self-report any inappropriate content.

Answer

Increase the temperature to 1.0 to ensure more diverse responses.

Question 5

What is the primary security risk of using an LLM to automatically generate and execute shell commands?

Accepted Answer

The model can be manipulated to execute unauthorized system commands.. The primary risk is 'Remote Code Execution' (RCE). If an LLM is given the agency to execute commands based on potentially malicious user input, an attacker can craft a prompt that tricks the model into running arbitrary, harmful code on the host system. This bypasses traditional security boundaries and allows the attacker to compromise the infrastructure, access sensitive files, or exfiltrate data from the environment.

Answer

The model will consume too many API tokens.

Answer

The model might hallucinate and generate incorrect commands.

Answer

The model will be unable to access the local file system.

Question 6

Refer to the exhibit. This input is an example of what type of security threat?

Accepted Answer

Prompt Injection.. This is a classic 'Prompt Injection' attack, specifically a 'jailbreak' attempt. The attacker is trying to override the developer's system instructions by using a command ('Ignore all previous instructions') to force the model to behave in a way that violates its original programming. Recognizing this pattern is essential for developers to build robust filters and guardrails that detect and reject these malicious attempts at model manipulation.

Answer

Data Exfiltration.

Answer

Denial of Service.

Answer

Cross-Site Scripting (XSS).

Question 7

Your team is building an agentic workflow that interacts with internal databases. Which TWO security practices should be implemented to prevent prompt injection attacks that could lead to unauthorized data exfiltration?

Accepted Answer

Implement a strict allow-list of tools and functions the model can execute.. Preventing prompt injection requires a defense-in-depth approach that separates user input from system instructions and enforces strict operational boundaries. By limiting the agent's capability to only necessary functions and using structured output formats, developers minimize the attack surface. These practices are critical because agentic workflows are highly susceptible to malicious instructions that override original system prompts, potentially leading to unauthorized data queries or unintended execution of dangerous internal operations.

Answer

Use hardcoded system prompts that are strictly enforced via fine-tuning.

Answer

Perform all API calls using a public-facing read-only database user.

Answer

Require human-in-the-loop approval for all model responses.

Question 8

A financial firm needs to identify potential jailbreak attempts against their Claude-powered chatbot. Which approach provides the most effective real-time detection?

Accepted Answer

Deploy a dedicated classification model to evaluate the safety of prompts and responses.. Real-time detection of jailbreak attempts is best achieved through a secondary, lightweight LLM or a specialized classifier that analyzes both the user input and the model's generated output. By evaluating the interaction for adversarial patterns—such as attempts to bypass safety filters—the security layer can intercept malicious prompts before they are fully processed or block harmful responses, providing a critical safety net that static rules cannot match.

Answer

Use a static keyword-based filter on the user input.

Answer

Audit the logs daily to search for suspicious query patterns.

Answer

Disable the ability for the model to access external URLs.

Question 9

What is the primary security benefit of using the Anthropic API in a Virtual Private Cloud (VPC) environment with a Private Link?

Accepted Answer

It ensures that traffic remains within a private network path, avoiding the public internet.. Using a Private Link connects your VPC directly to the API service over a private network connection, bypassing the public internet. This significantly reduces the attack surface, protects data from man-in-the-middle attacks, and ensures that traffic remains within the provider's backbone. It is a fundamental architectural requirement for enterprises that must adhere to strict regulatory compliance regarding data isolation and network security.

Answer

It increases the throughput of API requests.

Answer

It eliminates the need for API keys entirely.

Answer

It automatically encrypts the model responses locally.

Question 10

When designing a system that uses Claude to assist in writing code, what is the most important security consideration regarding the model's output?

Accepted Answer

Generated code must pass static analysis and human review before deployment.. AI-generated code may contain vulnerabilities, insecure patterns, or outdated libraries. The most critical security step is treating model output as untrusted input that requires rigorous automated scanning and human verification. This protects the organization from inadvertently deploying vulnerable code, which could lead to systemic security breaches, data loss, or non-compliance with secure software development lifecycle (SDLC) standards.

Answer

The model should be allowed to execute the code it generates.

Answer

The model should have access to your private repository for code quality.

Answer

The model's temperature should be set to 1.0 for better code creativity.

Question 11

Which TWO of the following are effective ways to protect sensitive data when building a RAG (Retrieval-Augmented Generation) pipeline?

Accepted Answer

Apply attribute-based access control (ABAC) to the vector database queries.. Protecting data in a RAG pipeline requires securing both the retrieval process and the generation process. Access control at the retrieval stage ensures that users only query information they are authorized to see. Simultaneously, data minimization during the augmentation phase prevents sensitive information from being unnecessarily exposed to the model. These steps are crucial because RAG systems often bridge the gap between secure databases and generative models, creating a high-risk surface for accidental data leakage.

Answer

Index all documents regardless of user access permissions.

Answer

Include the entire enterprise document store in the prompt context.

Answer

Store all retrieved documents in plain text within the application code.

Question 12

Your organization is integrating Claude for sensitive internal human resources queries. What is the most effective way to ensure the model does not reveal employee salary information?

Accepted Answer

Implement a data-masking middleware that filters sensitive fields from the retrieved context.. The most secure way to prevent sensitive data disclosure is to prevent the model from ever 'seeing' that data during the generation phase. By implementing a logic layer that filters the information based on the user's identity before it reaches the model, you ensure that the model is only provided with authorized context. This prevents the model from acting as an unauthorized channel for sensitive PII or payroll data.

Answer

Instruct the model in the system prompt to never discuss salaries.

Answer

Require the model to perform a sanity check on its own output for PII.

Answer

Encrypt all employee salaries using a salt and hash.

Question 13

Refer to the exhibit. An internal tool is configured to send user input directly to the API. Which security improvement should be applied to the architecture?

Accepted Answer

Implement a middleware to sanitize and block dangerous system commands in the user input.. The architecture is currently wide open to dangerous instruction execution. The application must include an input-validation or intent-classification layer before the request reaches the LLM. By checking if the request involves sensitive or destructive actions—such as file system deletion—the tool can block the request entirely, ensuring the model is never used to generate commands that could cause catastrophic system damage.

Answer

Increase the 'max_tokens' to ensure the deletion script is generated fully.

Answer

Use a more advanced model for the same request.

Answer

Add a disclaimer in the system prompt that deleting files is prohibited.

Question 14

A large-scale deployment of Claude is causing intermittent spikes in latency. Which security-related monitoring practice helps differentiate between a DDoS attack and legitimate heavy usage?

Accepted Answer

Analyze the request frequency and prompt diversity from specific origin IPs.. Differentiating between DDoS attacks and high legitimate load requires deep visibility into API request patterns. By tracking the distribution of source IPs, request frequency per token, and the complexity of prompts, security teams can identify anomalous patterns that signify an attack. This is crucial for maintaining availability without blocking legitimate users, ensuring that security measures are proportionate to the threat, rather than causing self-inflicted denial of service.

Answer

Check the total cost of API usage in the Anthropic dashboard.

Answer

Limit all users to one request per minute.

Answer

Disable all external access to the API immediately.

Question 15

What is the most secure method for handling long-term memory for an AI agent that handles sensitive customer data?

Accepted Answer

Store indexed, encrypted embeddings in a hardened vector database with fine-grained access control.. Long-term memory must be treated with the same security rigor as production databases. By using an encrypted, access-controlled vector store that limits retrieval to authorized queries, you ensure that sensitive information is stored safely and retrieved only for appropriate contexts. This avoids storing data in plaintext or in the model's context window, both of which are high-risk practices for managing sensitive information in agentic applications.

Answer

Save the entire conversation history as a text file in the local file system.

Answer

Keep all memory in the model's context window for the duration of the session.

Answer

Have the model generate a summary of the data and store only the summary.