Skip to main content

Security & compliance

Enterprise trust starts with evidence.

Evaluate the security boundary around your Amazon Bedrock AgentCore application. Review the controls in Kam State, the AWS configuration that needs verification, and our path toward SOC 2.

SOC 2 roadmap

Planned · no report available

Our goal is an independent SOC 2 Type II report for a defined Kam service boundary.

Kam AI has not completed a SOC 2 examination. An auditor, examination period, and completion date have not been announced.

Status reviewed

Explore the roadmap →
On this page

Security foundation

Review what the product enforces.

Kam records application-supplied outcomes and exact input references. Your application controls model calls, tool permissions, and external actions.

Source and local evidence · 2026-10-14

Principal and scope

The server resolves the workspace and scope authorized for the authenticated principal. A browser-supplied tenant identifier grants no access.

Stale authority

Serving and conditional-commit paths recheck lifecycle and authority. Revoked or expired authority and stale fences block serving or commit; retained bytes do not renew permission.

Missing required evidence

Exact reads preserve unavailable, expired, and unresolved positions. Missing required evidence blocks a current State selection or complete Context package; no latest substitution.

Retention and historical access

An earlier revision remains inspectable while retained and authorized. Logical expiry stops access before background TTL cleanup; a historical read grants no permission to act.

These are source and offline test boundaries. Deployed IAM, revocation, tenant isolation, erasure, and restore journeys still require cloud qualification.

Review security contracts and test boundaries →

AWS control requirements

Infrastructure is one part of the review.

AWS secures its cloud infrastructure. Kam must configure and operate its service controls; customers govern their applications, identities, and data.

AWS SOC reports cover AWS services within their stated scope. They do not provide a SOC 2 report for Kam AI.

The controls below are requirements for a qualified deployment, not a claim that these AWS services or configurations are enabled today.

Planned verification

Identity & tenant boundaries

Least-privilege IAM roles, scoped authorization, and administrative MFA.

Evidence to verify

Verify Cognito identity, role trust and permission policies, cross-tenant denial, revocation, access reviews, and joiner/leaver records.

Planned verification

Encryption & secrets

TLS in transit, encryption at rest, and restricted key and secret access.

Evidence to verify

Verify DynamoDB and any in-scope S3 encryption settings, KMS key policies, secret storage and rotation, and redaction from logs and model inputs.

Planned verification

Network & storage access

Explicit ingress, egress, resource policies, and private access where required.

Evidence to verify

Review public endpoints and bucket exposure, applicable VPC endpoints, resource policies, and environment separation. Record each justified exception.

Planned verification

Monitoring & audit evidence

Trace administrative changes, access decisions, and security findings.

Evidence to verify

Verify CloudTrail coverage, CloudWatch alerts, applicable AWS Config rules and threat detection, protected log retention, and documented response ownership.

Planned verification

Retention & recovery

Document deletion, backup retention, recovery targets, and restore tests.

Evidence to verify

Qualify DynamoDB point-in-time recovery, restore fences, and deletion across primary data, replicas, logs, and backups. Record RPO/RTO test results and exceptions.

Planned verification

Change & incident management

Review releases, remediate vulnerabilities, and rehearse incident response.

Evidence to verify

Retain change approvals, dependency scans, scoped security testing, remediation records, incident exercises, vendor reviews, and staff security training.

Path to SOC 2

From control design to independent assurance.

Security is the baseline category. Availability, confidentiality, processing integrity, and privacy will be scoped with customers and the independent auditor.

  1. Define the boundary

    Inventory the service, data flows, vendors, and risks. Assign control owners and agree the examination scope.

    Planned
  2. Close readiness gaps

    Document policies and qualify AWS access, encryption, isolation, monitoring, recovery, and incident procedures.

    Planned
  3. Operate & collect evidence

    Run the controls over an agreed observation period. Retain reviews, logs, exercises, changes, and remediation evidence.

    Planned
  4. Complete independent examination

    Engage an independent CPA firm, address findings, and share the resulting report subject to its distribution terms.

    Planned

This is a planned program, not an audit in progress or a guaranteed completion date. Technical tests alone do not establish SOC 2 assurance.

What do Type I and Type II mean?

Type I examines control design at a specified date. Type II also examines operating effectiveness over a specified period. A Type I report may be a readiness milestone; it is not a prerequisite to Type II.

Enterprise review

Bring your security requirements.

Discuss your deployment boundary, data classification, retention, and procurement requirements before an evaluation.

What to include in your request
  • Architecture, data flows, tenant access, and the evidence available for the reviewed release.
  • Required AWS regions, residency, subprocessors, data handling, and contract or DPA requirements.
  • Audit deadlines, penetration-test expectations, recovery targets, SSO needs, and incident notification terms.

Evidence availability and contractual commitments must be agreed for the evaluation. No SOC 2 report, production SLA, SSO capability, or residency commitment is implied.

Does SOC 2 certify an AI agent’s decisions?

No. SOC 2 examines controls within a defined system boundary. Kam’s retained history can support investigation, but it does not prove real-world truth, model safety, or permission to take an action.