TATER Security Trust Center

Data residency & encryption

AI safety posture

The AI Compliance Analyst, Evidence Agent, and 75-tool MCP server are powered by Anthropic Claude exclusively. We do not use OpenAI, Google Gemini, or any other foundation model.

What we don't do

• No training on customer data. Anthropic's commercial terms exclude API traffic from model training. We do not log prompts/responses to any third-party service beyond Anthropic's standard inference path.
• No cross-tenant prompt context. Every AI invocation is scoped to a single tenant; the MCP server enforces tenantId on every Cosmos query.
• No silent writes. Tools that change tenant state (create_risk_acceptance, execute_script, trigger_remediation, upsert_config_doc) require Admin role and are blocked for unverified MCP API key sessions until identity is verified via email-code.

What you control

• BYOK (Bring Your Own Key) — supply your own Anthropic API key per organization. Encrypted at rest with AES-256-GCM. AI features off by default; toggle from Settings → Features.
• Channel attribution — every AI-driven write is tagged in the audit log with via: copilot, claude, mcp, web, agent, or api. Filter Activity Log to see exactly what an AI agent did.
• Audit log integrity — set the AUDIT_SIGNING_KEY environment variable to enable HMAC-SHA256 per-entry signatures (CLAUDE.md §12ad). Verify offline with any HMAC tool.

Certifications & attestations

Full attestation evidence is shared under NDA during procurement. We do not publish auditor reports publicly.

Subprocessors

We use the minimum number of third-party services required to deliver the platform. Each subprocessor is contractually bound to confidentiality and security commitments comparable to ours.

Subscribe to subprocessor change notices via trust@tatersecurity.com. We provide 30 days advance notice before adding a new subprocessor that processes customer data.

Endpoint agent integrity

• SHA256 verification — every agent binary released is hashed at build time and the SHA256 is published via GET /api/agent/version (no auth required). The agent self-verifies updates against this hash before installing.
• Windows MSI — distributed via the same SHA256-verified channel as the binaries. Authenticode signing is on the roadmap; SHA256 verification of every binary is current.
• Single static binary on Windows / Linux (CGO-free, Go 1.26+) — no DLLs, no runtime dependency on a separate .NET or PowerShell module installer.
• Encrypted local config — DPAPI on Windows (CurrentUser scope), 0600 file perms with optional encryption on Linux/macOS.
• SBOM on request — software bill of materials for the Go agent and API can be generated on request via CycloneDX. Email security@tatersecurity.com.
• Phase 2 remote control — every WebRTC session requires explicit end-user consent on the endpoint (PE-3, 30-second fail-closed dialog). Capabilities (input injection, clipboard, recording) are individually toggleable; off by default at every level. Session recording (AU-14) is opt-in, encrypted at rest, 30-day TTL.

Audit log integrity

• Every create/update/delete on every TATER entity fires an AuditLog entry with: actor (UPN, OID), action, entity type, entity ID, timestamp (millisecond resolution), and channel (via: web/mcp/copilot/claude/agent/api/cron).
• Tamper-evident — set AUDIT_SIGNING_KEY env var to enable HMAC-SHA256 per-entry signatures. Canonical payload: id|tenantId|action|entityType|entityId|userId|timestamp. Verify with openssl dgst -sha256 -hmac KEY.
• SIEM forwarding — syslog (RFC 5424 CEF format) or HMAC-signed webhook (X-TATER-Signature: sha256=...). Compatible with Sentinel, QRadar, ArcSight, Splunk, Datadog, Sumo Logic, and more.
• CEF injection defense — pipe, backslash, and newline characters in user-supplied audit fields are escaped before CEF construction.

API security & access control

• Defense in depth — every write endpoint requires Admin role, every cross-org operation requires SuperAdmin, every API request is rate-limited (general: 100/min, write: 60/min, settings: 120/min, auth: 10/min). Per-IP and per-tenant buckets.
• IDOR protection — every PUT/DELETE handler verifies record ownership against the caller's organizationId before acting. Unbound API keys cannot use x-organization-id header to escalate scope (CLAUDE.md §12bp/§12bq).
• Cross-partition queries — explicit { forceQueryPlan: true } on every cross-partition Cosmos query so the SDK fans out across all tenants instead of silently routing to a single partition.
• SSRF prevention — every webhook destination is validated against an allowlist of public IPs; localhost, link-local, ULA, and private ranges are blocked by urlValidation.ts.
• CSP — strict Content Security Policy with frame-ancestors 'none', object-src 'none'. script-src limited to self + alcdn + jsdelivr + unpkg (MSAL CDN fallback chain).

Vulnerability disclosure

Security researchers and customers can report potential vulnerabilities to security@tatersecurity.com. We acknowledge within 1 business day and aim to triage within 5 business days.

• Safe harbor — good-faith research is welcome. Please do not run automated scans against our production infrastructure without coordination; reach out first.
• Severity tiers — Critical/High vulnerabilities receive a same-day acknowledgment and we work to a fix within 7 days. Medium within 30 days. Low at our discretion.
• Disclosure — coordinated disclosure preferred. Public advisories are issued after a fix is deployed; we credit researchers (with permission) in the advisory.
• No bug bounty at this time, but we send TATER swag and named credit for valid reports.

SLA & uptime

Data handling & retention

• What we collect — only what is needed to provide the service: scan results (compliance posture), control catalog (mostly default + per-org overlays), audit log entries, optional configuration documentation authored by the customer.
• What we don't collect — file contents from scanned endpoints, user passwords (Entra ID handles auth; we never see credentials), browsing history, application telemetry beyond the explicit pageview beacon (90-day TTL, opt-out via browser DNT or cookie).
• Retention — audit log entries: indefinite by default, configurable. Telemetry: 90 days. Session recordings (Phase 2 remote control): 30 days. Evidence agent jobs: 30 days. Customer can request deletion at any time.
• Deletion — within 30 days of contract termination we hard-delete tenant data from primary storage. Backups age out within the 7-day continuous backup window after the primary delete.
• Data export — every container is exportable via API; CSV/JSON download from each module page. POAM exports OMB A-130 / DoD eMASS .xlsx; SSP exports OSCAL JSON or Word .docx.

Contact