← Back to standardcompute.com
Standard ComputeSecurity
Security overviewUpdated October 2026

Security

How Standard Compute protects customer data, where it is processed, and how company administrators control which providers can handle it.

1. The service

Standard Compute is an AI gateway. Customers send OpenAI- or Anthropic-compatible API requests to one endpoint. For each request, we choose the model and inference provider that best fit the customer's cost, quality, speed and policy requirements.

Standard Compute AS is a Norwegian private limited company (org. no. 938 498 806), registered in Oslo, Norway.

2. How a request flows

Your application
Your agent, IDE or backend calls our API over TLS 1.2+.
Gateway
Authenticates, applies your policy, selects model and provider. Runs on Google Cloud in Finland (EU). An optional US gateway runs in the United States.
Routing
Forwards only to endpoints allowed by your policy.
Inference
An approved provider endpoint in your permitted region.

Your policy (allowed regions and allowed providers) is enforced on every step, including the request classifier, retries and fallbacks. If no permitted route is available, the request fails. It is never sent outside the policy. For workspaces restricted to Europe, the full request path stays in Europe; on eu.stdcmpt.com, TLS terminates in Finland. See Sub-processors.

3. What we store, and what we do not

DataHandlingLocation
Prompts and responsesNot stored. Processed in memory for the lifetime of the request only. Never used for training. There is no debug mode that logs content.Transient only
Request metadataTime, key ID, model, serving provider, token counts, cost and status. Used for billing, usage reporting and audit.EU (Frankfurt); operational logs in Google Cloud
Account dataName, email, company and team structure, settings. API keys are stored encrypted.EU (Frankfurt)
Payment dataHandled by Stripe. We never store card numbers.Stripe

4. Customer controls

ControlWhat it does
Region rulesRestrict inference to Europe, the United States or both, at company, team or key level. Restricted requests only use endpoints in the selected regions, including fallbacks.
Zero data retention enforcementRestricted workspaces only use endpoints classified as no-training and zero data retention. Response storage is disabled.
Provider allow-listAdmins approve a named list of inference providers. New providers never join the list without an admin decision. Applies to every request, classifier call and fallback.
Teams, roles and budgetsCompany, team and key hierarchy with spending limits. Lower levels can only narrow the rules set above them.
Admin audit logRecords administrative changes to members, keys, budgets and compliance settings.
Request provider logPer-request record of model, serving provider, region rule and outcome, including the fallback attempts providers report. Exportable. Kept for 12 months on Enterprise and the SLA add-on. No content.

Company admins set these rules on the Compliance page in the dashboard. Rules are opt-in. Keys without a rule use standard routing under each provider's own terms.

5. Technical safeguards

Encryption

All traffic uses TLS 1.2 or higher. Plain HTTP is refused, not redirected. API keys are encrypted at rest with AES-256-GCM and shown to the user once. Requests are validated by HMAC-SHA256 with constant-time comparison. Data at rest in our database and cloud infrastructure is encrypted by the hosting providers.

Tenant isolation

Row-level security in the database scopes every query to the authenticated account or company. Company settings can only be changed through authorised, versioned server-side operations.

Fail-closed policy enforcement

If the gateway cannot load or verify a customer's policy, the request is rejected rather than routed without it. Upstream providers are reached with dedicated service credentials. Credentials for one provider cannot be used for another.

Provider vetting

Every provider endpoint is reviewed before it can carry restricted traffic. See Provider Requirements.

6. Incident response

We notify affected customers of a personal data breach without undue delay, and in any event within 72 hours of becoming aware of it. We share the information available at the time and provide updates as the investigation progresses.

Vulnerability reports can be sent to contact@standardcompute.com. We acknowledge them within 2 business days and give an initial assessment within 7 business days. We will not take legal action against researchers who report in good faith and do not access other users' data or disrupt the service.

7. Assurance and certifications

FrameworkPositionStatus
GDPR / UK GDPRData Processing Agreement available. Account data hosted in the EU.In place
ISO 27001Information security management system built on ISO 27001:2022 controls, aiming for certification.In progress
SOC 2 Type IITo follow ISO 27001, based on customer demand.Planned
Infrastructure and inference partnersGoogle Cloud, AWS and Microsoft Azure hold ISO 27001 certification and SOC 2 reports for the services we use, both for hosting our gateway and for European inference. Mistral AI holds ISO 27001 certification and a SOC 2 Type II report. Supabase holds a SOC 2 Type II report.In place

Related: Data Processing Agreement · Sub-processors · Provider Requirements. Our completed answers to the LLM gateway security questionnaire are ready to paste into your vendor review, and we are happy to complete your own supplier security questionnaire.