Silk Compute Trust Center
Silk Compute maintains a security program to protect customer data. Here you can find our compliance work, security controls, subprocessors and security documents.
Infrastructure security
| Control | Status |
|---|---|
| Production application access restricted System access to production is restricted to authorized administrators. | In place |
| Production database access restricted Privileged access to the production database is restricted to authorized administrators. Application services reach it through restricted file permissions. | In place |
| Unique production database authentication enforced Access to production datastores requires a unique SSH key. | In place |
| Encryption key access restricted Encryption keys and production secrets are readable only by authorized administrators and the one service that needs each of them. | In place |
| Firewall access restricted Privileged access to the firewall is restricted to authorized administrators. | In place |
| Production OS access restricted Privileged access to the operating system is restricted to authorized administrators. | In place |
| Production network access restricted Privileged access to the production network is restricted to authorized administrators. | In place |
| Unique network system authentication enforced Access to production servers requires a unique SSH key. Password logins and remote root logins are disabled. | In place |
| Network firewalls utilized The server firewall denies inbound traffic by default. Public web traffic reaches production only through Cloudflare's network. | In place |
| Network segmentation implemented Application services listen only on the local interface behind a reverse proxy, and each service runs as its own system user. | In place |
| Network and system hardening standards maintained Services run sandboxed with reduced privileges, read-only code and resource limits, from hardening configuration kept in version control. | In place |
| Denial-of-service protection utilized Cloudflare protects every public endpoint against denial-of-service attacks. | In place |
| Intrusion detection system utilized Repeated failed SSH logins are blocked automatically. A daily integrity check alerts on changes to accounts, SSH, system services, scheduled tasks, listening ports and system packages. | In place |
| Log management utilized Web server and service logs are collected in the system journal and used to investigate abuse and errors. | In place |
| Infrastructure performance monitored An external monitor checks availability every five minutes and alerts on outages and missed health reports. | In place |
| Service infrastructure maintained Security updates for the operating system and server software install automatically every day. | In place |
| Backups performed and encrypted The database is backed up every day, encrypted, to separate cloud storage, and each backup is kept for 30 days. | In place |
| Production deployment access restricted Only authorized administrators can deploy to production. | In place |
| DNS security extensions enabled The domain is signed with DNSSEC, so resolvers can verify DNS answers. | In place |
Organizational security
| Control | Status |
|---|---|
| Unique account authentication enforced Every administrator signs in with a named personal account. The shared root account cannot sign in remotely. | In place |
| Administrator MFA enforced The admin dashboard requires a one-time code from an authenticator app, bound to the signed-in session. | In place |
| Administrator actions logged Every change made from the admin dashboard is recorded in an audit log. | In place |
| Information security policies established Written policies cover access control, change management, incident response, vulnerability management, data retention, encryption, vendors and workstations, and are reviewed every year. | In place |
| Security alerts reviewed Monitoring and integrity alerts go to the team inbox for review. | In place |
Product security
| Control | Status |
|---|---|
| Data transmission encrypted All connections use TLS, and browsers are instructed to use HTTPS only (HSTS). | In place |
| Data encryption utilized The production database is encrypted at rest with SQLCipher (AES-256). | In place |
| Support data encrypted at rest Support tickets and their attachments are encrypted in the database with AES-256-GCM. | In place |
| Credentials stored as hashes Passwords and API keys are stored only as hashes. An API key is shown once, when it is created. | In place |
| Email verification required Every account confirms its email address before it can sign in. | In place |
| Password policy enforced Account passwords must be at least 8 characters long. | In place |
| Customer MFA available Customers can turn on two-factor sign-in with an authenticator app, with one-time backup codes. | In place |
| Rate limiting enforced Sign-in, sign-up and password resets are rate limited, and addresses that fail API key checks repeatedly are blocked for a time. | In place |
| Bot protection utilized Sign-ups are checked for bots before an account is created. | In place |
| Request size limits enforced Oversized requests are rejected before they reach the application. | In place |
| Security headers enforced Pages send a content security policy, block framing by other sites, and disable content type sniffing. | In place |
| API key spending limits available Customers can set a spending limit on any API key and revoke a key at any time. | In place |
Internal security procedures
| Control | Status |
|---|---|
| Change management procedures enforced Every code change is tracked in version control, and production ships only from the main branch. | In place |
| Configuration management system established Server configuration, including the firewall, SSH, services and web server, is kept in version control and installed through reviewed scripts. | In place |
| Database changes backed up A database backup is taken before every schema change. | In place |
| Penetration testing performed An internal penetration test of the platform and API was completed in November 2026. | In place |
| Incident response plan established A written incident response plan covers detection, severity, containment, customer notification and review. | In place |
| Continuity and Disaster Recovery plans established A written plan covers restoring the service from encrypted off-site backups after the loss of the database or the server. | In place |
| Continuity and Disaster Recovery plans tested Restoring from backup is tested every quarter. | In place |
| Vulnerability disclosure process established Anyone can report a vulnerability by email, encrypted with our published PGP key. | In place |
| Incident notification committed If a security incident affects customer data, we notify the affected customers. | In place |
Data and privacy
| Control | Status |
|---|---|
| Prompts and outputs not stored Prompts and model outputs pass through our servers and are not stored. | In place |
| Customer data not used for training Silk Compute never uses prompts or model outputs to train models. | In place |
| Customer data deleted upon leaving Deleting an account removes the account, API keys, usage and payment history from the database right away. | In place |
| Data retention procedures established For each request we keep only billing records: the model, token counts, cost, time and API key. Server logs are removed automatically as they age. | In place |
| Payment card data not stored Card payments are processed by Stripe. Card numbers never reach our servers. | In place |
| No data sale or advertising trackers We never sell customer data, and there are no analytics or advertising cookies. | In place |