What is attack success rate in AI red teaming, and what are its limits as a metric?
AI security · Pro 2
Short answer
Attack success rate is the share of attack attempts that reached their goal: if 12 of 400 jailbreak prompts got a harmful answer, it is 3%. It is useful for tracking versions on the same suite, by category. But it depends on the attacks chosen and how success is judged, and a low rate does not prove safety.
Deeper answer
Report it per category (prompt injection, data leakage, harmful content, tool misuse) and per severity, because one critical leak matters more than many mild failures. Keep the attack suite and the judge fixed when comparing versions, or the numbers are not comparable. Check the judge itself: an LLM judge should be compared with human labels on a sample.
Limits: attackers adapt and are not limited to your suite; a model can overfit to known attacks; multi-turn and indirect attacks are often under-tested; and non-determinism means repeating each attack several times gives a fairer picture. Pair the rate with severity, coverage of the threat model and time-to-fix for findings.
What a strong answer covers
- Defines it: successful attempts divided by total attempts
- Reports by category and severity
- Keeps suite and judge fixed to compare versions
- Validates the judge and repeats attacks
- Limits: suite choice, adaptive attackers, low rate is not safe
Answer it out loud in the mock interview about 1.5 min
Follow-ups to expect
- Why should each attack be run more than once?
- How would you check that your LLM judge is reliable?
Common mistakes
- Comparing rates from different attack suites or judges
- Treating a low rate as proof that the system is secure
Learn it
Security tests in CI for AI apps
- Attackers now use LLMs to write phishing emails. What changes for defenders, and what still works?
- Threat model an LLM chatbot for a retail bank that can answer questions and show a customer's recent transactions.
- What is the CIA triad, and how would you use it to judge a security problem?
- What is data poisoning, and how would you protect a model you train or fine-tune from it?
- As a security engineer, how would you protect a company's finance team from deepfake voice or video calls from the 'CEO'?
- You find code that builds a SQL query by joining strings with user input. What is the risk and how do you fix it?
- What is the difference between hashing and encryption, and which do you use for passwords?
- An employee's laptop shows signs of malware at 11 pm. Walk me through your incident response.
- Explain least privilege. How would you apply it to a new team's cloud and app access?
- Your SOC wants an LLM copilot to summarise alerts and suggest actions. What are the risks and how do you manage them?
- What is tool poisoning in MCP or agent tools, and how do you defend an agent against it?
- What is MITRE ATT&CK, and how does a detection team actually use it?
- A vendor says their ML intrusion detector is 99% accurate. Why might it still drown your SOC in false alarms?
- What is a model extraction attack, and how would you defend a model you serve through an API?
- Walk me through the OWASP Top 10 for LLM applications. Which risks would you check first in a RAG chatbot?
- How would you defend a company against phishing? Where do SPF, DKIM and DMARC fit in?
- A teammate wants to load a model file they downloaded from a public hub. What supply chain risks do you check?
- You are asked to red-team your company's customer support chatbot before launch. How do you go about it?
- Tell me about a security incident or near miss you handled. What did you do and what changed afterwards?
- Staff are pasting company data into public AI tools without approval. How would you handle shadow AI?
- Your SOC gets thousands of SIEM alerts a day and analysts are ignoring them. How do you fix alert fatigue?