Trust and security

Security · Compliance · Data handling · AI use policy · Subprocessors · Vulnerability disclosure · Status · Verify a receipt

Security

• Tenant isolation with Postgres row-level security on every customer table.
• Encryption in transit (TLS) and at rest. Request payloads are encrypted again with a per-organization data key that rotates monthly.
• Every decision is written to a signed, hash-chained ledger. Signing keys live in Google Cloud KMS, and anyone can verify an export offline. Anyone can also verify a receipt in the browser, and follow it to the anchored ledger entry behind it.
• Customer-managed keys for sealed context: agents encrypt sensitive fields with your own key, and only your codec endpoint can open them.
• Single sign-on with OpenID Connect (OIDC), and SCIM 2.0 provisioning, for Business and Enterprise plans, set up once in PromptReports ID with no third-party identity vendor. People removed in your directory lose access within 5 minutes.
• Roles and permissions: owner, admin, policy editor, approver, auditor and developer.
• IP allowlists for the dashboard, API and MCP server, and PII masking on approval cards.
• High-risk decisions need a passkey or one-time code from the last 5 minutes.
• Audit log export: the full ledger as JSONL or CSV, and evidence packs, from the dashboard or the API.
• Controls on AI agent actions are the product: approval policies, scopes on API keys, rate and spend limits, and a kill switch per action type.

Compliance

We are preparing for a SOC 2 Type 1 audit with an independent auditor and will publish the report period here once it is complete. We do not hold ISO 27001 yet.

Data processing agreement: email privacy@promptcompute.ai and we send our DPA for signature.

Data handling

• No training. We do not use customer data, agent requests, decisions or questionnaire content to train AI models, ours or our providers'. Our agreement with Anthropic forbids training on our API traffic.
• Retention. Request payloads, including AI inputs such as agent context and questionnaire drafts, are kept for 30 days on Free, 365 days on Team and 7 years on Business and Enterprise, then destroyed by deleting their monthly key. Legal holds keep a held request until released. Scanned page text is kept 90 days.
• Data residency. Organizations choose the United States or the European Union when they are created. EU organizations' data and jobs stay on EU infrastructure.
• Deletion. Owners can ask for organization deletion at privacy@promptcompute.ai.

AI use policy

We use AI in two places. Readiness scores use Claude to find verbatim quotes on public pages; every quote is checked against the page, and low-confidence results wait for a person. Questionnaire answers are drafted by Claude from your own evidence and documents with citations, and each one stays a draft until someone on your team approves it. AI never decides an approval request: people do, or policies you wrote.

Subprocessors

We give customers 30 days' notice before adding a subprocessor. Email privacy@promptcompute.ai to get notices.

Vulnerability disclosure

Report security issues to security@promptcompute.ai. We reply within 2 business days, will not take legal action against good-faith research that avoids privacy harm and service disruption, and credit reporters who want it. Our security.txt has the details.