LLM Prompt Injection Awareness & Mitigation Practices Survey

Welcome! Thank you for participating in this survey about your experiences with LLM application development and security. Your participation is entirely voluntary, and you may stop at any time. There are no right or wrong answers — we are interested in your honest perspectives and practices. All responses are confidential and will be reported only in aggregate. This survey takes approximately 6–8 minutes to complete.

Which area best describes your primary role?

• Back-end engineering
• Front-end engineering
• Full-stack engineering
• Machine learning / data science
• DevOps / SRE
• Security engineering
• QA / Test automation
• Other (please specify)

How confident are you in your personal understanding of prompt injection risks and mitigations?

Which of the following threat vectors do you consider when designing or reviewing LLM features? Select all that apply.

• Prompt injection (malicious instructions in user input)
• Indirect prompt injection via external content sources
• Jailbreaks / model override of system policies
• Data exfiltration or leakage via prompts or responses
• Tool misuse or over-permissioned agent actions
• Training data poisoning
• Prompt leakage or system prompt version exposure
• None of these

Which of the following prompt injection mitigations has your team adopted? Select all that apply.

• Input validation and sanitization of user prompts
• System prompt hardening (e.g., instruction hierarchy, delimiters)
• Output filtering or content safety checks
• Role-based access controls for LLM tool/action permissions
• Monitoring and logging of LLM interactions
• Canary tokens or honeypots in system prompts
• Sandboxing or isolation of LLM execution environments
• None of these
• Not sure

In the last 6 months, have you encountered suspected prompt injection or jailbreak activity in your systems?

• Yes — confirmed incident
• Possibly — suspicious behavior, not confirmed
• No

Rank the following metrics by how much you prioritize them when evaluating prompt injection mitigations (top = highest priority).

1. False positive rate (legitimate inputs incorrectly blocked)
2. Detection rate (malicious inputs correctly caught)
3. Latency impact on user experience
4. Ease of implementation and maintenance
5. Coverage across attack types

We'd like to explore your experience with LLM security practices in a bit more depth. An AI moderator will ask a couple of follow-up questions based on your earlier responses.

How many years of professional software development experience do you have?

• Less than 1 year
• 1–3 years
• 4–6 years
• 7–10 years
• More than 10 years

Thank you for completing this survey! Your responses will contribute to a better understanding of LLM security practices across the developer community and help improve secure development guidance.

Which of the following best describes your current involvement with LLM-integrated features?

• I currently build or maintain LLM-integrated features
• I am planning to integrate LLMs in the next 6 months
• I am not currently working with LLMs

How well-prepared is your team or organization to defend against prompt injection attacks on your LLM-integrated applications?

Rank the following LLM security concerns from highest to lowest priority for your work.

1. Direct prompt injection via user input
2. Indirect prompt injection via external content
3. Data leakage via prompts or responses
4. Over-permissive tool or agent actions
5. Model override / jailbreak to bypass policies

How often does your team evaluate for prompt injection or jailbreak risks?

• Weekly or more often
• Every 2–3 weeks
• Monthly
• Quarterly
• Less often or never

Briefly describe the incident and how it was handled. Please omit any sensitive or proprietary information.

What is your biggest obstacle to managing prompt injection risk today?

Based on your responses in this survey, please share any additional thoughts or experiences related to LLM security and prompt injection that we haven't covered. (Optional)

Approximately how many employees are in your organization?

• 1–10
• 11–50
• 51–200
• 201–1,000
• 1,001–5,000
• 5,001+

Where have you learned about prompt injection risks and mitigations? Select all that apply.

• Vendor or framework documentation
• OWASP Top 10 for LLM Applications
• Academic papers or preprints
• Security blogs or newsletters
• Conference talks or workshops
• Internal training or peer guidance
• Social media or forums
• I have not sought out information on this topic

Which methods does your team use to test for prompt injection vulnerabilities? Select all that apply.

• Adversarial red teaming by engineers
• Automated evaluation suites or checklists
• Unit or integration tests for prompts
• Canary or honeytoken detection
• Shadow deployment with monitoring and alerts
• External penetration testing
• We do not currently test for this

Where are you primarily located?

• Africa
• Asia
• Europe
• Latin America
• Middle East
• North America
• Oceania

What is your team's default response policy when LLM input may be unsafe?

• Allow but sanitize or validate content
• Block and ask the user for clarification
• Escalate to human review
• Varies by context or risk level
• Not sure

Which industry best describes your organization?

• Technology / Software
• Finance / Banking
• Healthcare
• Retail / E-commerce
• Manufacturing
• Education
• Government / Nonprofit
• Other (please specify)

Which types of tools or platforms does your team use to mitigate prompt injection risks? Select all that apply.

• LLM gateway or proxy with policy enforcement
• Content moderation or safety APIs
• Vector database with filtering or access controls
• Open-source guardrails libraries (e.g., Guardrails AI, NeMo Guardrails)
• Cloud provider built-in safety features
• Custom internal middleware or services
• None