SOC 2 Type II
Independent controls attestation
AWS
Cloud foundation for the platform
Governed AI
Engines decide, AI explains
IaC
Infrastructure as code, reviewed & repeatable
At a glance
A short list for diligence. Deeper architecture is below.
SOC 2 Type II
Independent attestation of control operating effectiveness over a period.
Encryption & isolation
Data protected in transit and at rest; workloads designed for least privilege.
Governed AI
Engines produce the numbers, the AI explains them, and judgment calls go to a person.
Evidence trail
Logging and supervision so examiners can ask what happened — and get an answer.
How the program stays ready between exams
Practices that support SOC 2 Type II for a financial advice platform include ongoing monitoring, evidence you can hand over, and a clear path when a questionnaire arrives. The program is built so drift can be fixed before diligence asks, rather than only during annual audit season.
Continuous control monitoring
Security and compliance controls are monitored on an ongoing cadence — not only during annual audit season — so drift can be spotted and remediated before diligence asks.
Evidence readiness
Control evidence is collected and organized so customer diligence, security questionnaires, and auditor requests can move faster without ad-hoc screenshot hunts.
Formal policies
Information security, access, and operational policies are maintained as living documents that map to SOC 2 control expectations — reviewed as the product and infra evolve.
Access reviews
Employee and operator access is subject to periodic review so privileges stay aligned with role — especially important across client, reporting, and Mission Control surfaces.
Vendor & subprocessors
Critical vendors that touch infrastructure, observability, or customer data are tracked through a vendor-risk process appropriate for a SOC 2 Type II program.
Audit cadence
Independent SOC 2 Type II examination covers the operating effectiveness of controls over a period — the standard RIA and enterprise buyers expect for financial platforms.
Encrypt, scope, and keep secrets out of code
Encryption and access patterns stay consistent in transit, at rest, and across service roles, so a diligence questionnaire does not get a different story in each environment.
Encryption in transit
Client and API traffic is designed to use TLS (TLS 1.2+) at the edge. Sensitive service-to-service paths stay inside the AWS network where practical.
Encryption at rest
Core data stores and object storage are encrypted at rest. Sensitive reporting tables use customer-managed KMS keys with key rotation enabled.
Secrets management
Operational secrets are designed to live in AWS Secrets Manager — retrieved by scoped service roles, not hard-coded into application images.
Least-privilege access
Infrastructure and application roles follow least-privilege IAM patterns so services receive only the permissions required for their job.
AWS, designed as code
Athena’s production posture sits on Amazon Web Services with CloudFormation-managed infrastructure so network, identity, data, and AI controls can be reviewed and reproduced across stages.
Isolated network design
VPC · private subnets · multi-AZ. Application workloads run in Amazon VPC with private subnets across availability zones. Services are designed to run behind private load balancing — not as publicly exposed task hosts.
Hardened edge
CloudFront · WAF · TLS. Public surfaces terminate TLS at the edge. AWS WAF managed rule sets help filter common web exploits before traffic reaches application layers to block malicious traffic and safeguard core infrastructure.
Repeatable infrastructure
CloudFormation · Stacker. Environments are provisioned with AWS CloudFormation (infrastructure as code) so network, identity, data, and AI controls can be reviewed, versioned, and reproduced across stages.
Observable operations
CloudTrail · CloudWatch · Datadog. API activity monitoring, service health alarms, and application observability are part of the operating model — so unusual changes and production issues can be investigated quickly.
Leading models, with the engines in charge
Athena AI uses leading models from OpenAI and xAI to explain answers in your firm’s voice. The model never decides the recommendation: Athena’s planning and investing engines produce the numbers from the household’s own data, the AI explains them, and anything that needs judgment goes to a person on your team. It’s the same split described on Athena AI and in the planning methodology guide.
Athena does not use your firm’s or your clients’ data to train AI models. It is used only to answer your clients and your team.
Where a firm needs the AI isolated, Athena can run it on Amazon Bedrock with Bedrock Guardrails inside its own AWS infrastructure, as it does for American Century’s rollover assistant.
| Athena AI | How |
|---|---|
| Numbers and recommendations | From the engines |
| Explanation in your voice | From the model |
| Every conversation | Kept on the household |
| Judgment calls | Escalated to a person |
- Athena AI
- Numbers and recommendations
- How
- From the engines
- Athena AI
- Explanation in your voice
- How
- From the model
- Athena AI
- Every conversation
- How
- Kept on the household
- Athena AI
- Judgment calls
- How
- Escalated to a person
For a diligence reader, what matters most is reconstructability: what the client asked, what the engines proposed, what the model explained, and what a person approved or sent back in Mission Control. Athena AI is not a free-form advice black box bolted on beside your book. It sits on the household record next to overnight investing, advice-triggered communication, and the compliance gate for acquisition creative.
Identity, roles, and humans in the loop
Operator identity uses Amazon Cognito with MFA options on client and advisor pools, short-lived tokens, and group and scope packages so Mission Control, reporting, and tooling can be granted by duty rather than as a flat admin model. Product supervision keeps exceptions and approvals visible on the household, not buried in a side spreadsheet. Cash requests, trade exceptions, and creative review show up where an operator can approve, change, or send back without rebuilding the file from a blotter and a separate CRM.
| Mission Control | Approvals |
|---|---|
| Cash requests | Needs you |
| Trade exceptions | In queue |
| Creative review | Gated |
| Operator sign-in | Cognito · MFA |
- Mission Control
- Cash requests
- Approvals
- Needs you
- Mission Control
- Trade exceptions
- Approvals
- In queue
- Mission Control
- Creative review
- Approvals
- Gated
- Mission Control
- Operator sign-in
- Approvals
- Cognito · MFA
Mission Control is your household CRM, so the conversation record, the overnight investing exceptions, and the advice-triggered messages an examiner asks about sit on one reconstructable household. Schwab remains the custodian for the accounts your firm manages on Athena, and your books and records stay where they are. Ads on Facebook, Instagram, TikTok, Google, LinkedIn, and YouTube still clear a compliance gate before publishing, and the firm remains responsible for marketing posture.
Built for advice data, custody context, and firm operators
Household PII, account context, and AUM-related signals are treated as regulated-practice data, not generic SaaS telemetry. Design favors isolation, scoped access, and reconstructable activity so a reviewer can follow what a client saw and when. Advisor and operations surfaces use Cognito-backed identity with group and OAuth-scope packages so reporting, Mission Control, and tooling access can be granted by duty, not as a flat admin model. Mission Control and reporting paths are designed with domain and package isolation so operators see only the households they are authorized for, and fail closed when tenancy cannot be resolved. Critical DynamoDB tables enable point-in-time recovery; object storage for user uploads uses server-side encryption and blocks public bucket access by default.
Financial data sensitivity
Household PII, account context, and AUM-related signals are treated as regulated-practice data — not generic SaaS telemetry. Design choices favor isolation, scoped access, and reconstructable activity.
Role-based operator access
Advisor and operations surfaces use Cognito-backed identity with group and OAuth-scope packages — so reporting, Mission Control, and tooling access can be granted by duty, not as a flat admin model.
Tenant-aware reporting walls
Mission Control and reporting paths are designed with domain and package isolation so operators see only the households they are authorized for — fail-closed when tenancy cannot be resolved.
Recovery readiness
Critical DynamoDB tables enable point-in-time recovery. Object storage for user uploads uses server-side encryption and blocks public bucket access by default.
What buyers should ask any agentic platform
Athena is designed so those answers live in the product and the compliance program, not only in a slide deck after the demo.
Where do humans still approve?
Can you reconstruct what a client saw and when?
How are PII and account data encrypted and accessed?
How does marketing creative clear compliance?
What happens when the model is wrong?
Is the platform SOC 2 Type II compliant?