Security Telemetry Model

Principles

• Data minimization: default to metadata only (hashes, counts, timings). Payload capture is opt-in and scoped.

• Selective retention: rolling windows (e.g., 7–30 days) locally; proofs live on-chain, not content.

• Zero-trust transport: mTLS between agent↔engine; all records signed at source; replay-proof envelopes.

• Workspace isolation: per-tenant keys; strict boundary enforcement for multi-org deployments.

PII handling

• No credential contents, messages, or file bodies by default.

• User identifiers can be tokenized (e.g., reversible inside tenant, irreversible externally).

• Support for DLP guardrails to block exfil of sensitive classes outright.

---

2.5 Consensus & Event Ledger (how proofs live)

Lifecycle

1. Emit: AI Engine signs event_hash.

2. Batch: micro-batch proofs for efficiency; attach Merkle root.

3. Anchor: submit to L1; receive tx/slot reference.

4. Expose: Dashboard & API show verifiable link; auditors can check independently.

5. Evolve: subsequent actions (quarantine released, creds rotated) can append follow-up proofs forming an immutable incident chain.

Verifier API (read-only)

GET /v1/proofs/{event_hash}
→ { anchor: { chain, tx, slot }, timestamp, class, ai_confidence, merkle_path }

External SIEM/SOAR

• Webhooks push incident + proof.

• Splunk/Sentinel apps can verify inclusion with Merkle proofs client-side.