Improve Prompt Injection and Jailbreak Detection Accuracy

[coderabbitai]

🧠 Detection Accuracy Improvement

Description

Improve the core detection engine's accuracy for identifying prompt injection and jailbreak attempts, reducing both false positives and false negatives.

Tasks

• Audit current detection logic and identify common false positive patterns
• Research and incorporate latest prompt injection attack vectors (2024/2025)
• Improve detection for multi-turn jailbreak attempts
• Add detection for indirect prompt injection attacks
• Tune confidence scoring thresholds
• Benchmark detection accuracy against a labeled dataset
• Document detection methodology and limitations

Acceptance Criteria

• Reduction in false positive rate (document baseline first)
• Coverage of major jailbreak categories
• All changes backed by test cases

Difficulty: 🔴 Hard / 🔴 Critical

Labels: ai/ml enhancement critical SSoC26

The core mission of TENET — make the detection engine smarter and more robust!