<!-- Generated from the canonical HTML page. Do not edit separately. -->
# Security - Architecture, Controls & Disclosure | Ankra

Canonical: https://ankra.ai/security

How Ankra is built: an outbound-only agent, scoped access and configurable agent autonomy, and a controls matrix that only lists what is shipped, opt-in, or in progress.

Security

# Easier operations. Explicit boundaries.

Understand Ankra’s architecture, the controls available today and how to secure the applications and infrastructure you run.

[Read the documentation](https://docs.ankra.ai/security/overview)  [Talk to our team](mailto:hello@ankra.ai)

Architecture

## Where Ankra sits in your trust boundary

A SaaS control plane, an outbound-only agent inside your cluster, and Git as the source of truth between them.

1. SaaS control plane

  ### The control plane runs with us

  The UI, API, and orchestration run at platform.ankra.app by default. It holds metadata, encrypted credentials in Vault, and the audit log - not your workloads. Enterprise customers can run the control plane self-hosted instead.
2. In-cluster agent

  ### The agent runs with you, and only dials out

  One agent per cluster connects outbound over NATS with per-agent JWT credentials and rotating session tokens. Nothing connects inbound - your API server is never exposed to us.
3. Git

  ### Git is the source of truth

  Desired state lives in your Git repositories. The platform reconciles clusters to Git, so turning Ankra off leaves your clusters running from history you own.

platform.ankra.app/agents

Agent connection  prod-eu · connected

- direction  outbound only
- transport  NATS
- identity  per-agent JWT credentials
- session  rotating session tokens
- inbound  none

The agent dials out. Your cluster accepts nothing in.

### The honest part: the agent runs with cluster-admin

The agent has cluster-admin inside your cluster - it has to, because it manages anything you ask it to, from namespaces to CRDs. We would rather tell you that plainly than bury it. What constrains that power is the operating model around it:

- Scoped roles and configurable agent autonomy - gated proposals wait for approval
- Drift guards - sync windows and prune protection constrain out-of-band change
- Lease serialisation - one operation at a time per cluster, no racing changes
- Execution records and agent transcripts for the work performed

Controls

## The controls matrix, as it stands today

Everything below is shipped, opt-in, or actively in progress. There is no aspirational tier - if it is not listed, we have not built it.

### Identity & access

 4 controls

[SSO login with MFA](https://docs.ankra.ai/platform/account-security)

TOTP, WebAuthn passkeys, and recovery codes on every account.

 Shipped

[Organisation-level MFA requirement](https://docs.ankra.ai/platform/account-security)

Require MFA for every member of your organisation, enforced at login.

 Shipped

[Role-based access control](https://docs.ankra.ai/guides/roles-and-access)

Built-in and custom roles, with assignments scoped to cluster groups.

 Shipped

[Short-lived kubeconfig cluster access](https://docs.ankra.ai/guides/cluster-access)

SSO-backed cluster credentials that expire on their own, with instant revoke-all.

 Shipped

### API & AI surface

 5 controls

[API tokens hashed at rest](https://docs.ankra.ai/reference/tokens)

SHA-256 hashes only - secret material lives in Vault, with a 90-day default expiry.

 Shipped

[MCP scope isolation](https://docs.ankra.ai/platform/mcp-server)

mcp:read and mcp:write are separate scopes; MCP tokens are rejected on the REST API and vice versa.

 Shipped

[MCP OAuth 2.1 with PKCE only](https://docs.ankra.ai/platform/mcp-server)

Standards-based authorisation per RFC 9728, RFC 8414, and RFC 7591.

 Shipped

[Always-on AI secret redaction](https://docs.ankra.ai/platform/mcp-server)

Pattern and entropy detection runs before anything is persisted.

 Shipped

[Secret values never exposed to AI tools](https://docs.ankra.ai/platform/mcp-server)

AI tools see key names and counts only, never the values.

 Shipped

### Secrets & data

 3 controls

[Vault-backed secret storage](https://docs.ankra.ai/security/data-handling)

Platform secrets are stored in HashiCorp Vault or OpenBao.

 Shipped

[End-to-end secret encryption](https://docs.ankra.ai/guides/sops)

SOPS with AGE, using per-organisation keys.

 Opt-in

[In-cluster SOPS decryption](https://docs.ankra.ai/guides/sops)

Encrypted secrets are decrypted inside your cluster - plaintext never transits the platform.

 Shipped

### Audit & governance

 6 controls

[Append-only audit log](https://docs.ankra.ai/guides/audit-log)

Every administrative change is recorded in a log that can only grow.

 Shipped

[Organisation audit viewer](https://docs.ankra.ai/guides/audit-log)

Search and filter the full audit history from the platform.

 Shipped

[Audit log export](https://docs.ankra.ai/guides/audit-log)

CSV and JSON export of audit events.

 In progress

[12-month standard audit retention](https://docs.ankra.ai/guides/audit-log)

Audit events retained for twelve months as standard.

 In progress

[Draft, approve, commit](https://docs.ankra.ai/security/overview)

Every AI-proposed change is human-approved and lands as an attributed Git commit.

 Shipped

[Drift detection](https://docs.ankra.ai/security/overview)

Sync windows and prune guards catch and constrain out-of-band change.

 Shipped

### Cluster & infrastructure

 4 controls

[Hardened in-cluster agent](https://docs.ankra.ai/security/overview)

Non-root, read-only filesystem, all capabilities dropped, seccomp enabled.

 Shipped

[Outbound-only connectivity](https://docs.ankra.ai/security/overview)

Per-agent credentials and no inbound access to your cluster.

 Shipped

[Always-on log redaction](https://docs.ankra.ai/security/overview)

Logs are redacted at the source, before they leave your cluster.

 Shipped

[Hardened provisioning](https://docs.ankra.ai/security/hardening)

Scoped cloud firewalls, private networking, bastion hosts, Vault-managed SSH keys, and 15-minute join tokens.

 Shipped

### Vulnerability management

 8 controls

[Security Center](https://docs.ankra.ai/guides/cluster-security)

Fleet-wide findings, per-cluster posture, and remediation priorities, powered by Trivy Operator.

 Opt-in

[NetworkPolicy over-privilege scoring](https://docs.ankra.ai/guides/cluster-security)

Ingress and egress scored per cluster: unprotected workloads, over-broad rules, and what to tighten.

 Opt-in

[Pod-security policy violations](https://docs.ankra.ai/guides/cluster-security)

Kyverno baseline results per cluster, installed in audit mode with an explicit switch to enforce.

 Opt-in

[One-click security baseline install](https://docs.ankra.ai/guides/cluster-security)

Installs pinned Trivy Operator and Kyverno on a cluster from a single action.

 Shipped

[Security baseline on new clusters](https://docs.ankra.ai/guides/cluster-security)

A security baseline installed by default on newly created clusters.

 In progress

[Accepted-risk dispositions](https://docs.ankra.ai/guides/cluster-security)

Accept a finding with an expiry date, and revoke the acceptance at any time.

 Shipped

[Daily posture snapshots and trends](https://docs.ankra.ai/guides/cluster-security)

Posture captured daily, so you see direction rather than a point in time.

 Shipped

[Scheduled security reports](https://docs.ankra.ai/guides/security-reports)

Recurring reports by email, with CSV export.

 Shipped

### Compliance

 3 controls

[Benchmark compliance posture](https://docs.ankra.ai/guides/cluster-security)

CIS and NSA/CISA pass rates per cluster, plus Trivy configuration audit as a second axis.

 Opt-in

SOC 2 programme

An ISMS is being built against the Trust Services Criteria. Audit in progress.

 In progress

[Compliance evidence export](https://docs.ankra.ai/guides/cluster-security)

CSV or JSON over a date window: posture history, dispositions with reasons, audit summary, and attestations.

 Shipped

Your clusters

## The other half: securing what you run

Everything above is how we secure the platform. The Security Center is the product side of that, covering the clusters you run on it.

Image vulnerabilities, benchmark compliance, pod-security policy violations, and NetworkPolicy over-privilege across every cluster in the organisation. Findings become decisions you record once, with a reason and a review date, rather than a list that returns unchanged on the next scan.

[Explore the Security Center](https://ankra.ai/security/security-center)

Disclosure

## Found something? Tell us.

We take reports from security researchers seriously and we answer them ourselves.

### Report a vulnerability

Email  [security@ankra.ai](mailto:security@ankra.ai) with what you found and how to reproduce it. We acknowledge reports within three business days, and good-faith research will not be met with legal action.

[Read the disclosure policy](https://ankra.ai/security/disclosure)

### Buying Ankra?

Procurement and compliance material lives on our trust page: our SOC 2 posture, what data the control plane holds, and the sub-processors behind the platform. Product detail on every control lives in the [documentation](https://docs.ankra.ai/) .

[Visit the trust page](https://ankra.ai/trust)

Your next chapter

## You built it.
*Let’s get it live.*

Give your application a path to production.
 Give your team a simpler way to run it.

[Get your app live](https://platform.ankra.app/auth/signup)  [Try a playground](https://ankra.ai/playground)

First 30 vCPU of platform usage included.
Try it without a card; a card unlocks full access and €59 of credit.

From something you built
to something people use.

- Deploy from your repository
- Run on infrastructure you choose
- Operate with your team and agents