Three AI Models Scored 100% on a CISSP-Style Exam Experiment

Summary

An author tested three AI models on 150 self-written CISSP-style questions across all eight domains. Each scored 150/150; the author argues that exam performance does not mean AI can make cybersecurity decisions in uncertain real-world situations.

Key points

• The author wrote 150 questions across all eight CISSP domains and did not use real exam questions or a paid question bank.
• The experiment had 100 standard questions and 50 complex questions requiring selection of the best answer among multiple technically correct options.
• ChatGPT, Gemini, and Claude each scored 150 out of 150, with the author judging their reasoning correct on every question.
• The author notes that the questions were self-written, so personal bias may have influenced the results.
• The author argues that passing a recognition-based exam does not establish that AI can create cybersecurity strategies or make decisions with limited information.
• The article recommends that security professionals use AI for suitable tasks while focusing on work where human judgment is valuable.

People

Products