Trust & security

Built for work that can't leave the building.

Generic AI can't touch privileged or regulated data. Miko can — because privacy is architectural. Inference runs inside your control boundary, nothing privileged egresses, and every request is logged and reviewable.

The controls

Privacy enforced by architecture, not policy.

  1. 01

    Data boundaries

    Your workload runs in a single-tenant dedicated environment — your organization's alone, isolated per engagement, under customer-held encryption keys. Egress is denied by architecture: anything that can't be served inside the boundary is refused, not forwarded.

    • No training on your data — ever
    • Processing stays in-region, under your residency policy
    • Scoped, audited access — no standing access to your data
    • Encrypted in transit and at rest throughout the pipeline
    • Retention you control; deletion is honored
  2. 02

    Model routing

    A pre-inference gateway classifies every request before any model sees it. Privileged and regulated classes are served inside your boundary by open-weight models we run — no third-party model vendor in the chain for that class. A request that can't be served privately is refused, not forwarded.

  3. 03

    Human approval gates

    Consequential actions wait for a named human: drafts are drafted, never sent, until someone with authority signs off. Approval chains are defined per engagement with named approvers, and the system cannot bypass or talk its way past its own gates.

  4. 04

    Audit logs

    Every classification, routing decision, draft, and approval lands in an append-only, tamper-evident log. The log is yours to receive — engagements end with evidence, not anecdotes.

  5. 05

    Evals & red-teaming

    Every workflow carries a versioned eval suite with a regression set, re-run before any material change ships — a failure that reappears blocks the release. Workflows touching sensitive data are adversarially tested on their extraction and drafting paths: prompt injection, boundary bypass, approval-gate evasion, exfiltration.

  6. 06

    Shadow mode

    New workflows run alongside your team with zero external effect before they run with any. Output is compared against what your people actually did — performance is proven on real work before the system is trusted with it.

  7. 07

    Incident response

    Every workflow has a tested kill switch a human can throw without the system's cooperation, and a documented, tested rollback path. If an incident affects your data, we notify you without undue delay.

  8. 08

    Certification roadmap

    Frameworks

    • SOC 2 Type II
    • ISO 27001
    • HIPAA / HITECH

    Architecture built to meet these standards — formal certification in progress.

    Formal certification is in progress. The architecture is built to meet SOC 2, ISO 27001, HIPAA / HITECH, and in-region residency from the start, and we will walk your security team through the controls, our current status, and a DPA or BAA under NDA.

Security review

Built for your security team.

  • We complete security questionnaires (SIG, CAIQ) and your own.
  • Architecture diagrams, data-flow maps, and our reports — shared under NDA.
  • We sign your DPA and, where applicable, a BAA.
  • Penetration-test summaries and our security package on request.

Walk your security team through the controls.

Architecture, current certification status, and a sample deliverable — under NDA, candidly.

Book a briefing →