Correlate the incident
Bring approved metadata from AI systems, agents, guardrails, cloud audit, vector databases, observability, and security tools into one investigation.
Briard-AI security add-on
When AI behaves unexpectedly, preserve what happened.
KAIDAN is Briard-AI's incident-response add-on for AI and LLM environments. It correlates approved metadata into an evidence-cited investigation record without silently becoming an inline enforcement control.
Metadata-only by default Evidence cited Tenant scoped
What KAIDAN is
Incident forensics for the AI layer.
Briard-AI answers how an AI system is governed. KAIDAN answers what the available evidence says happened during an AI or LLM incident. Briard owns the customer, subscription, identity, and add-on authorization experience; KAIDAN remains authoritative for its evidence ledger, cases, findings, connector receipts, and custody verification.
What it does
Move from scattered signals to a defensible investigation.
Separate evidence from inference
Keep observed facts, source assertions, detector inferences, reviewer conclusions, gaps, and contradictions visibly distinct.
Preserve verifiable custody
Retain tenant-scoped, tamper-evident evidence and produce export packages whose integrity can be checked independently.
How it works
Sources stay attributable at every step.
- 01
Connect approved sources
A customer operator selects permitted metadata sources and grants least-privilege access.
- 02
Correlate with receipts
KAIDAN preserves source attribution and leaves missing external receipts explicitly unknown.
- 03
Investigate with context
Responders review timelines, roles, risks, evidence citations, gaps, and contradictions.
- 04
Export and verify
Teams create custody-verifiable packages for authorized review and offline integrity checking.
Who it is for
Teams responsible when AI and LLM systems become incidents.
Security and incident response
Reconstruct AI and LLM activity without treating an alert as proof.
AI platform and engineering
Give responders source context, connector health, and ownership without copying application code.
Risk, legal, and compliance
Review evidence-cited cases and acknowledged unknowns while specialists retain decision authority.
Honest boundaries
Evidence first. Claims stay disciplined.
- Normal intake is metadata-only; raw prompts, responses, files, and secrets do not belong there.
- Restricted forensic capture is separate, disabled by default, and requires its own controls and approvals.
- KAIDAN observes and preserves evidence. It does not silently block AI systems or make response decisions.
- Missing source receipts remain unknown; an alert or detector result is not presented as established fact.
- Customer deployment, provider credentials, external notarization, and independent review are separate readiness states.