STUDENT PROJECT
AI threat assessment — local & global
How should many weak observations become useful risk information without turning AI into an unquestioned authority?
The problem
Local nodes can observe concrete events. A wider system can correlate reports across networks, history and persistent unit identity. The research question is how to combine those views responsibly.
Possible work
Confidence scoring, explainability, adversarial manipulation, false-positive control, local-versus-global models, evidence provenance, model evaluation, human oversight and recovery when a unit's behavior improves.
Students can work from the current AI-assessment code or replace the approach entirely.
Discuss it with the TaraSec AI
Ask which evidence is available in the current system, how the assessment path works and where a new model or evaluation framework could plug in.