Cybersecurity capstone: hardened AI assistant

The brief

Every company is adding AI assistants, and few test them like attackers would. Build an assistant with access to documents and tools, red-team it against the OWASP Top 10 for LLM applications (prompt injection, data leaks, excessive agency), fix what you find, and present a security report. Good looks like: a documented attack plan, findings with evidence, fixes with before-and-after tests, and a presentation to a panel.

What you will build

1. Build an assistant with documents and tools
2. Plan attacks against the OWASP Top 10 for LLM applications
3. Run direct and indirect prompt injection tests
4. Fix findings and re-test
5. Add monitoring for abuse
6. Present the security report to a panel

Free, open data to build it with

• OWASP Top 10 for LLM applications ↗ OWASP
• MITRE ATT&CK ↗ MITRE

Check each dataset's licence before you publish your work. Your mentor gives your team its own version of the brief (data, scope or region), so every team's project is its own.