Skip to main content

SECURITY & TRUST

Built to hold.

Every claim on this page is drawn from the production infrastructure. Nothing here is aspirational. If it is on this page, it is running.

Infrastructure

Cloud provider

Amazon Web Services — us-east-1 exclusively. No other regions, no other clouds.

Encryption at rest

AES-256 via AWS KMS. Every database cluster, every S3 object. Separate KMS keys per data tier.

Encryption in transit

TLS 1.3 enforced at the load balancer. TLS 1.2 minimum. No plaintext permitted.

Compute

AWS ECS Fargate. No shared infrastructure. Tasks run in isolated network namespaces.

Secrets management

AWS Secrets Manager only. Zero API keys in code or environment variables. All secrets injected at task runtime.

Database

Aurora PostgreSQL 15 with deletion protection enabled and automated backups retained for 7 days.

Tenant isolation

Row-level security is enforced at the database layer on every table in carltonintel_db. Every query runs inside a user-scoped RLS context. No query can return rows belonging to another user. There are no exceptions to this rule and no code paths that bypass it.

IAM roles follow the principle of least privilege. The application runtime role has write access only to its own CloudWatch log groups and S3 prefix. No cross-database access. No production credentials in application code.

FOUNDER PRIVACY

The Anchor guarantee

The Anchor is CIA's founder-private intelligence lens. Its output is never shared, never exported, never surfaced in any channel, never aggregated, and never used to train any model.

This is not a privacy policy statement — it is an architectural constraint. Share and export controls are disabled at the component level for Anchor output. The Anchor table carries the same row-level security as every other table, scoped to the authenticated user's own session. No code path in the platform bypasses these controls.

A founder's honest self-assessment should not be available to anyone but that founder. CIA enforces this in code, not just policy.

AI provider continuity

CIA runs a documented multi-provider continuity architecture. In June 2026, a federal directive suspended a major AI model across all platforms for 19 days. CIA was designed with that failure mode in mind.

The primary inference path runs directly to Anthropic. The confirmed failover path runs through AWS Bedrock — response confirmed under 3 seconds in live testing. COOP continuity candidates are scored and ranked, ready to activate. An automated circuit breaker monitors provider health and routes to the backup path when a threshold is crossed.

Every inference path runs inside the same US-only AWS boundary. No model call leaves the us-east-1 perimeter.

Continuity planning is documented to NIST SP 800-53 CP-2 standards. Vendor oversight follows NIST SA-9 external services protocol. Annual continuity drills are scheduled as part of the operational calendar.

Compliance posture

SOC 2

SOC 2-aligned architecture. Evidence collection underway. Formal audit scheduled for the SOC 2 Type II window.

NIST 800-53

Controls mapped across AC, AU, SC, IA, and SI families. CP-2 contingency plan and SA-9 vendor oversight documents committed to the repository.

Access control

JWT authentication with 15-minute access tokens and 7-day refresh rotation. TOTP MFA available at the COMMAND tier. Session termination on logout.

Audit trail

Structured logging via CloudWatch with 90-day retention. AWS CloudTrail enabled with log validation. All CIA delivery events logged to cia_delivery_audits.

CIA is a commercial product. It does not seek FedRAMP authorization and is not deployed on GovCloud infrastructure. The NIST alignment described above supports SOC 2 evidence collection and operational resilience — not federal authorization.

Subscription and cancellation

Subscriptions can be cancelled at any time from within the platform. Cancellation takes effect at the end of the current billing period — access continues until that date. No penalty, no retention flow, no dark patterns.

Billing is processed by Stripe. CIA does not store payment card data. All transaction handling occurs within Stripe's PCI-compliant infrastructure.

Responsible disclosure

Security issues should be reported directly to [email protected] with "Security Disclosure" in the subject line. Reports are reviewed by Carlton A. James personally. Confirmed vulnerabilities receive written acknowledgment within 3 business days.

CIA is a commercial platform. Do not test against production systems or other users' accounts. Report findings before disclosing publicly.