Wired into every service, not bolted beside them
Explicit bridges connect the security plane to compute, the data graph, messaging, metrics and storage, with one unified health view across them, plus service-layer guards every service composes: the authorization gate, the mutual-TLS gate, the compartment resolver and the read-audit emitter.
Because the platform generates and processes the events itself, there is no ingestion delay, no schema mismatch and no normalization loss between the platform and its detection layer.
| Layer | What it carries |
|---|---|
| Cryptographic foundation | Key hierarchy with epochs, forward-secret ratchet, sealing, hardware storage and attestation, post-quantum primitives with hybrid constructions. |
| Identity and access | Federation, adaptive risk-based step-up, conditional access, and a three-layer permission catalog with a deny-by-default evaluator. |
| Transport | Mutual TLS between cells, inbound and outbound guards, an egress URL guard, a web application firewall, and a threat-level enforcer that tightens access when detection raises the posture. |
| Data security | The in-engine enforcement plane, a data-loss-prevention policy layer, masking including format-preserving encryption, data residency pinned at region, zone or rack with failover held inside the pinned boundary, and privacy engineering. |
| Threat detection | An open event schema, detection rules evaluated as continuous streaming queries, anomaly engines, kill-chain correlation and network-level detection. |
| Threat intelligence | A security knowledge graph, indicator lifecycle, technique mapping, campaign tracking, and a cross-cell sharing mesh. |
| Response | Playbooks, incident correlation, quarantine, deception, forensics with legal holds, and red and purple team with coverage scoring. |
| Advanced security | Model supply chain, deployment gating, runtime guards, adversarial robustness and output provenance. |
| Governance and evidence | Control libraries, maturity scoring, entitlement and privilege analysis, quantified risk, and a sealed evidence pack. |
| Through all nine | Hash-chained audit with gaps counted as first-class events, and bridges into compute, data, messaging, metrics and storage with one unified health view. |
Nine layers, each building on the one below
Cryptographic foundation
A key-encryption-key hierarchy with epochs and a persistent epoch record, a forward-secret ratchet, lifecycle management and migration with a background re-wrap worker. Cell, node and sealing identity keys; per-purpose key separation; high-sensitivity custody; and destruction of key material as a first-class operation.
Hardware and attestation
Key storage across TPM 2.0, PKCS#11 and hardware tokens, with sealed vaults, signing without exporting. Verification of TPM quotes, enclave documents and cloud attestation tokens against manufacturer roots, with a replay guard and an attestation audit log. Hardware evidence grades to a trust level rather than a boolean.
Identity and access
Password, multi-factor, WebAuthn, SAML 2.0, OIDC, SCIM, LDAP, API keys and device authentication; session hardening with device fingerprinting; adaptive risk-based step-up; conditional access by location, device and risk; and a three-layer permission catalog with a deny-by-default unified evaluator.
Transport
Mutual TLS between cells with identity carried in the client certificate, key pinning, inbound and outbound guards, an egress URL guard, a web application firewall, per-client and per-endpoint rate limiting, circuit breaking, and a threat-level enforcer that tightens access automatically when the platform's own detection pipeline raises the posture.
Data security
The in-engine enforcement plane, plus a data-loss-prevention policy layer with pattern and machine-learning content detectors and an inline interceptor, ten masking strategies including format-preserving encryption, classification through the highest levels with compartments, and data residency pinned at any level of the topology (region, zone or rack, plus the orthogonal cloud-provider dimension) re-resolved against live fabric membership on every placement and egress decision so failover never crosses the pinned boundary, and a required constraint no live cell can satisfy refuses rather than quietly degrading to the wrong jurisdiction. Privacy engineering is its own package, anonymisation, differential privacy with a budget, privacy impact assessment, and a query rewriter that enforces privacy at the query level.
Threat detection
An open event schema with normalizer, enrichment, ingest governor, persistent store, observable index and dead-letter queue; detection rules evaluated as continuous streaming queries rather than after-the-fact log search; statistical and machine-learning anomaly engines; multi-stage kill-chain correlation; and network-level detection normally found only in dedicated products, command-and-control, beaconing, domain-generation algorithms, tunnelling, fast flux and reconnaissance.
Threat intelligence
A security knowledge graph connecting indicators, entities, campaigns and techniques, with durable graph storage and relationship traversal; indicator lifecycle management; technique mapping; campaign tracking; and a cross-cell sharing mesh with transport selection, bandwidth management and a tactical journal for disconnected operation.
Response
Response playbooks with runs and per-run detail, incident correlation and management, quarantine, deception with honeypots and canary tokens, forensics with case workspaces, evidence collection and legal holds, red and purple team with a technique catalog and detection-coverage scoring, and service-level tracking on response.
Governance and evidence
Control libraries, evidence collection, maturity scoring, remediation planning and continuous monitoring; posture scoring with framework coverage; entitlement and privilege analysis with escalation-path detection; migration planning with total-cost modelling; quantified risk on an industry-standard model; and a sealed evidence pack assembled from the compliance report, drift findings and deployment profile.
We do not ask a security operations centre to abandon its tooling
Three postures, one platform, and our own telemetry arrives already normalised to an open schema, so feeding your existing stack costs nothing.
Feed it
We forward to your destinations behind a fail-closed gate, so a misconfigured forwarder stops rather than leaks. Our events arrive already normalised.
Correlate alongside it
We ingest from your tools and correlate at the event level across environments they cannot reach, a distributed, low-and-slow pattern is faint at any single vantage point and visible from many at once.
Replace it
The newest of the three postures, and we say so. Start with the first two.
The cross-environment correlation is the part with no equivalent: a cloud-bound or single-vantage detection stack cannot see a pattern that only resolves when observed from many places at once. Detections are also weighted by how reliable the observing cell has proven to be.
A supply chain for models, not a guardrail
This is the most under-sold part of the security story, and it is a story almost nobody else tells: software-bill-of-materials generation and dependency scanning for models, source validation, signed artifacts, model provenance, scanning, quarantine and revocation snapshots, an artifact registry with manifest handling, plus attestation and integrity verification.
On the deployment side: a deployment profile, a deployment gate, an approval broadcaster and a governed loader. At runtime: prompt guardrails with a guardrail audit, input validation, a classification gate, a safety filter on a read-down model, an execution sandbox and a source guard for executable content.
Federated training keeps the data still, each cell trains on its local shard and ships a cell-signed parameter update to a control-owner cell that aggregates a sample-weighted quorum.
What is actually built
3
certificate systems at operational parity. X.509, SSH and PGP each issue, sign, verify, revoke and rotate
5
PKI protocol servers so ordinary clients just work, revocation, workload identity and key management over standard protocols
8
inbound connector types, including industrial protocols, so a plant's sensors land in the same event plane as its endpoints
10
masking strategies, from redaction and tokenisation to format-preserving encryption
Alongside them: post-quantum primitives with hybrid classical constructions for key exchange, PGP, SSH and X.509, and the hybrid X.509 issuance path is built and reachable, minting a certificate that carries a classical key alongside an ML-DSA-65 post-quantum signature. It is opt-in per policy, defaults to classical, and fails closed: if the post-quantum path cannot complete it refuses rather than downgrading to a classical certificate. We describe it as built, opt-in and fail-closed, never as a shipped post-quantum PKI in production, and CNSA 2.0 remains a policy posture that is assessed, not certified. Beside it: algorithm agility with peer capability negotiation and a migration planner for moving material off a deprecated curve; threshold cryptography with key ceremonies and escrow under legal hold, so recovery needs a quorum rather than a single custodian; certificate transparency with full path validation; and cryptographic audit with external witnessing, so an outside party can hold evidence that the ledger was in a given state at a given time.