Skip to main content

Detector overview

AI.Sentinel ships with 55 built-in detectors across three categories:

CategoryCountPurpose
Security31Prompt injection, jailbreaks, PII / credential leakage, covert channels, indirect injection, RAG poisoning
Hallucination9Phantom citations, fabricated authorities, contradictions, stale knowledge, confidence decay
Operational15Repetition loops, blank responses, truncated output, language switches, persona drift, sycophancy

Detector modes​

Every detector falls into one of three execution modes:

  • Rule-based — fast regex or heuristic. Always active. Sub-microsecond per call.
  • Semantic — uses embedding cosine similarity via IEmbeddingGenerator. Language-agnostic. No-op until opts.EmbeddingGenerator is configured.
  • LLM escalation — fires a second-pass LLM classifier. No-op until opts.EscalationClient is configured. Used for ambiguous or low-confidence rule-based hits.

Severity model​

Each detector returns a DetectionResult carrying a Severity (None, Low, Medium, High, Critical) and a reason string. The pipeline aggregates per-detector severities into a Threat Risk Score (0–100) that drives the Intervention Engine.

Detector ID convention​

Built-in detectors use three prefixes:

  • SEC-NN — security
  • HAL-NN — hallucination
  • OPS-NN — operational

Custom detectors authored via opts.AddDetector<T>() must use a different prefix to avoid collisions with future official detectors. Examples: ACME-01, MYORG-CUSTOM-01.

Tuning​

Every detector — built-in or custom — can be disabled or have its severity output clamped via opts.Configure<T>(c => ...). Floor and Cap apply only to firing results; Clean results pass through unchanged.

opts.Configure<WrongLanguageDetector>(c => c.Enabled = false);
opts.Configure<JailbreakDetector>(c => c.SeverityFloor = Severity.High);
opts.Configure<RepetitionLoopDetector>(c => c.SeverityCap = Severity.Low);

Where to next​