Skip to main content
Truefoundry is built with security and compliance as foundational principles. Our platform provides enterprise-grade security features, comprehensive compliance certifications, and robust data protection to ensure your AI workloads and data remain secure. For detailed security information, compliance reports, and trust documentation, visit trust.truefoundry.com.

​
Our Security Philosophy

Our security strategy is anchored in two core principles: defense in depth and secure by design. As an AI platform that orchestrates model deployments, handles MCP communications, and manages sensitive data flows across your infrastructure, security is not an afterthought—it’s embedded into every layer of our architecture. We believe modern software security requires proactive measures rather than reactive responses. Our approach combines industry-standard security practices with secure defaults and automated tooling to create a robust security posture that scales with your needs.

Defense in Depth

Multiple layers of security controls ensure that no single point of failure can compromise the system

Secure by Design

Security considerations are integrated from the earliest design stages, not bolted on later

Automated Security

Continuous automated scanning with 30+ security tools on every code change

Human Expertise

Expert security reviews and threat modeling complement automated tooling

​
Beyond Automated Tools: Human-Centered Security

While our automated security tooling provides comprehensive coverage, we recognize that true security requires human expertise and collaborative oversight. Our security posture extends beyond automated scanning:
Our security engineers conduct thorough code reviews focusing on logic flaws, business logic vulnerabilities, and complex attack vectors that automated tools might miss. Every significant change undergoes security-focused peer review.
Regular security assessments evaluate our attack surface, identify potential threat vectors, and validate our defense mechanisms against real-world attack scenarios. We continuously update our threat models as the landscape evolves.
All major design decisions undergo security architecture review to ensure security considerations are embedded from the earliest stages. This prevents costly security retrofits and ensures consistent protection.
Regular penetration testing by internal and external security experts validates our defenses and identifies potential weaknesses before they can be exploited.
We actively engage with the security research community, maintain responsible disclosure processes, and leverage collective intelligence to identify and address emerging threats promptly.
Developers across the organization receive security training and act as security advocates within their teams, creating a culture of security awareness that permeates every aspect of development.

​
Comprehensive Security Pipeline

Our security pipeline operates at multiple levels, ensuring that vulnerabilities are caught as early as possible in the development lifecycle:
1

Pre-commit Security Gates

Before any code reaches our repository, it must pass through rigorous pre-commit hooks including:
  • Static analysis with Bandit for common security issues
  • Semantic pattern matching with Semgrep for complex vulnerability patterns
  • Secrets detection with Dodgy and Gitleaks for hardcoded credentials
  • Type checking and code quality enforcement
2

Continuous Integration Security

Our CI/CD pipelines implement automated security scanning on every pull request and commit:
  • 30+ security scans trigger automatically on every PR
  • CodeQL and Semgrep for deep semantic code analysis
  • Gitleaks for comprehensive secret detection across git history
  • pip-audit and npm audit for dependency vulnerability scanning
  • Container security assessment for all containerized components
3

Supply Chain Security

We maintain strict oversight of our software supply chain:
  • Automated dependency vulnerability scanning
  • Software Bill of Materials (SBOM) generation for complete transparency
  • License compliance checking to ensure all components meet security standards
  • Continuous monitoring of upstream dependencies for new vulnerabilities
4

Container Security Hardening

Our containerized deployments follow security best practices:
  • Multi-stage builds to minimize attack surface
  • Minimal base images with the latest security updates
  • Non-root user execution and read-only filesystems
  • Comprehensive scanning with Grype
5

Runtime Security Monitoring

Beyond build-time security, we implement runtime monitoring to detect and respond to potential threats in production environments.
This multi-layered approach ensures that security vulnerabilities are identified and addressed at the earliest possible stage, reducing risk and maintaining the integrity of our platform.

​
Compliance Certifications

Truefoundry maintains multiple compliance certifications to meet enterprise security and regulatory requirements:
  • SOC 2 Type II: Certified for security, availability, processing integrity, confidentiality, and privacy
  • GDPR: Compliant with the General Data Protection Regulation for data privacy
  • HIPAA: Compliant with Health Insurance Portability and Accountability Act for healthcare data

SOC 2 Type II Logo

GDPR Logo

HIPAA Logo

Our Compliance Certifications

Compliance certifications apply to Truefoundry’s managed infrastructure. For self-hosted deployments, compliance depends on your infrastructure and security controls.

​
Data Security

​
Data Residency and Sovereignty

If you are self-hosting the control-plane, Truefoundry’s architecture ensures that your data remains under your control:
  • Data Stays in Your Environment: When deployed on your infrastructure, all data, models, and artifacts remain within your cloud account or on-premises environment
  • No Data Egress: Deploying on your own infrastructure eliminates data egress costs and ensures data never leaves your environment
Truefoundry control-plane communicates with our central authentication/licensing server and exchanges the following information:
Truefoundry control-plane uses our remote authentication server to authenticate the users logging into the Truefoundry platform. If you have added SSO, this will be a chained authentication flow where in the user is validated by Truefoundry and your IdP both. To read more on this, please refer to our SSO documentation.
The control-plane communicates with our central analytics server (https://analytics.truefoundry.com) and exchanges the following information:
  • The number of clusters connected to the control-plane
  • The addons installed in each of the clusters along with their versions
  • The version of the control-plane
  • The number of requests flowing through the gateway (Will be enabled in future)
This is an example of a sample payload that is sent to the analytics server. This request is sent once every hour from the control-plane to the analytics server.
{
  "clusters": [
    {
      "clusterName": "test-cluster1",
      "kubernetesVersion": "v1.33.4-eks-e386d34",
      "provider": "aws",
      "tenantName": "tenant1",
      "controlPlaneVersion": "0.90.0",
      "controlPlaneURL": "https://app.truefoundry.com",
      "truefoundryControlPlane": true,
      "controlPlaneProvider": "aws",
      "isConnected": true,
      "addons": [
        {
          "name": "truefoundry",
          "version": "0.90.0",
          "healthStatus": "Healthy",
          "syncStatus": "OutOfSync",
          "sourceNamespace": "argocd",
          "destinationNamespace": "truefoundry",
          "addonName": "truefoundry"
        },
        {
          "name": "tfy-gpu-operator",
          "version": "0.4.0",
          "healthStatus": "Healthy",
          "syncStatus": "OutOfSync",
          "sourceNamespace": "argocd",
          "destinationNamespace": "tfy-gpu-operator",
          "addonName": "tfy-gpu-operator"
        },
        {
          "name": "prometheus",
          "version": "69.6.0",
          "healthStatus": "Healthy",
          "syncStatus": "Synced",
          "sourceNamespace": "argocd",
          "destinationNamespace": "prometheus",
          "addonName": "prometheus"
        },
        {
          "name": "keda",
          "version": "2.17.1",
          "healthStatus": "Healthy",
          "syncStatus": "Synced",
          "sourceNamespace": "argocd",
          "destinationNamespace": "keda",
          "addonName": "keda"
        },
        {
          "name": "argocd",
          "version": "7.8.26",
          "healthStatus": "Healthy",
          "syncStatus": "Synced",
          "sourceNamespace": "argocd",
          "destinationNamespace": "argocd",
          "addonName": "argocd"
        },
        {
          "name": "argo-rollout",
          "version": "2.39.5",
          "healthStatus": "Healthy",
          "syncStatus": "Synced",
          "sourceNamespace": "argocd",
          "destinationNamespace": "argo-rollouts",
          "addonName": "argo-rollouts"
        },
        {
          "name": "aws-efs-csi-driver",
          "version": "3.1.8",
          "healthStatus": "Healthy",
          "syncStatus": "Synced",
          "sourceNamespace": "aws-efs-csi-driver",
          "destinationNamespace": "aws-efs-csi-driver",
          "addonName": "aws-efs-csi-driver"
        },
        {
          "name": "tfy-logs",
          "version": "0.1.10",
          "healthStatus": "Healthy",
          "syncStatus": "OutOfSync",
          "sourceNamespace": "argocd",
          "destinationNamespace": "tfy-logs",
          "addonName": "tfy-logs"
        },
        {
          "name": "aws-ebs-csi-driver",
          "version": "2.41.0",
          "healthStatus": "Healthy",
          "syncStatus": "Synced",
          "sourceNamespace": "aws-ebs-csi-driver",
          "destinationNamespace": "aws-ebs-csi-driver",
          "addonName": "aws-ebs-csi-driver"
        },
        {
          "name": "metrics-server",
          "version": "3.12.1",
          "healthStatus": "Healthy",
          "syncStatus": "Synced",
          "sourceNamespace": "kube-system",
          "destinationNamespace": "kube-system",
          "addonName": "metrics-server"
        },
        {
          "name": "karpenter",
          "version": "0.5.6",
          "healthStatus": "Healthy",
          "syncStatus": "Synced",
          "sourceNamespace": "kube-system",
          "destinationNamespace": "kube-system",
          "addonName": "tfy-karpenter"
        },
        {
          "name": "tfy-prometheus-config",
          "version": "0.2.14",
          "healthStatus": "Healthy",
          "syncStatus": "OutOfSync",
          "sourceNamespace": "prometheus",
          "destinationNamespace": "prometheus",
          "addonName": "tfy-prometheus-config"
        },
        {
          "name": "argo-workflows",
          "version": "0.45.12",
          "healthStatus": "Healthy",
          "syncStatus": "Synced",
          "sourceNamespace": "argo-workflows",
          "destinationNamespace": "argo-workflows",
          "addonName": "argo-workflows"
        },
        {
          "name": "tfy-istio-ingress",
          "version": "0.3.2",
          "healthStatus": "Healthy",
          "syncStatus": "OutOfSync",
          "sourceNamespace": "istio-system",
          "destinationNamespace": "istio-system",
          "addonName": "tfy-istio-ingress"
        },
        {
          "name": "istio-base",
          "version": "1.25.4",
          "healthStatus": "Healthy",
          "syncStatus": "OutOfSync",
          "sourceNamespace": "istio-system",
          "destinationNamespace": "istio-system",
          "addonName": "istio-base"
        },
        {
          "name": "istio-discovery",
          "version": "1.25.4",
          "healthStatus": "Healthy",
          "syncStatus": "OutOfSync",
          "sourceNamespace": "istio-system",
          "destinationNamespace": "istio-system",
          "addonName": "istio-discovery"
        },
        {
          "name": "elasti",
          "version": "0.1.16",
          "healthStatus": "Healthy",
          "syncStatus": "Synced",
          "sourceNamespace": "elasti",
          "destinationNamespace": "elasti",
          "addonName": "elasti"
        }
      ],
      "autopilotStatistics": {
        "clusterName": "tfy-prod-euwe1",
        "tenantName": "internal",
        "autopilot": { "addons": false, "service": false },
        "p50CpuUsage": 4.49,
        "p50CpuRequested": 65.822,
        "p50CpuAllocatable": 211.02,
        "p50MemoryUsage": 133.447,
        "p50MemoryRequested": 255.933,
        "p50MemoryAllocatable": 649.102,
        "p50GpuCountUsage": 0,
        "p50GpuCountRequested": 0,
        "p50GpuCountProvisioned": 0,
        "p50GpuMemoryUsage": 0,
        "p50GpuMemoryProvisioned": 0
      }
    }
  ]
}
The control-plane communicates with our central catalogue server (https://catalogue.truefoundry.com) to get the public pricing of the instance types of AWS/GCP/Azure that are shown on the Truefoundry dashboard. We also fetch the public pricing of the LLM models that are added to the AI Gateway.
Truefoundry uses Sentry to collect crash and error logs from the control-plane and send it to Sentry’s servers. This helps us identify and fix issues quickly.

​
Encryption

  • Encryption at Rest: Encrypted at rest using AES-256 encryption
  • Encryption in Transit: All network communications use TLS 1.2 or higher to encrypt data in transit
  • Secret Management: Integration with cloud-native secret managers (AWS Secrets Manager, Azure Key Vault, GCP Secret Manager) for secure credential storage

​
Access Control and Authentication

Truefoundry supports multiple authentication mechanisms:
  • Single Sign-On (SSO): Integrate with your identity provider (IdP) including SAML 2.0 and OIDC-compatible providers like Google, Microsoft Entra ID, Keycloak, Okta, etc. Learn more about our SSO documentation.
  • JWT Token Authentication: API access using JWT tokens from your identity provider
  • API Keys: Secure API key-based authentication for programmatic access
  • Token Rotation: Truefoundry supports automated token rotation for the API keys and tokens used to authenticate with the Truefoundry platform. Learn more about our Token Rotation documentation.
Comprehensive role-based access control (RBAC) ensures users have appropriate permissions:
  • Tenant-Level Permissions: Control access at the organization level with Tenant Admin and Tenant Member roles
  • Resource-Level Permissions: Granular permissions for workspaces, clusters, ML repositories, secret groups, and provider accounts
  • Team-Based Access: Organize users into teams with shared permissions and access controls
  • Fine-Grained Permissions: Viewer, Editor, and Admin roles for different resources
For detailed information on access control, see Access Control.

​
Infrastructure Security

Truefoundry’s split-plane architecture provides security through separation, ensuring sensitive data and workloads remain under your control:Data Never Leaves Your Environment
  • The compute plane runs entirely within your Kubernetes cluster (EKS, GKE, AKS, OpenShift, or on-premises). Truefoundry does not provide or access compute resources directly—all workloads execute in your controlled environment
  • Your models, datasets, and artifacts stay within your cloud account or on-premises infrastructure, eliminating data egress risks
  • The computeplane load balancer can be put behing a private endpoint limiting all external access to the compute plane.
  • The compute plane doesn’t require any ingress from the control-plane. Its the compute plane that intiates the connection to the control-plane. The tfy-agent connects to the control plane via outbound-only secure WebSocket connections, meaning no inbound ports need to be opened in your firewall
  • ControlPlane can be selfhosted within your VPC in our enterprise plan.
  • Private Network Deployment: Deploy control plane within your VPC for complete network isolation
  • TLS/SSL Encryption: All API communications encrypted with TLS 1.2+
  • Firewall Integration: Compatible with your existing firewall and network security policies
  • VPC Isolation: Support for deployment within private VPCs and on-premises networks
  • Image Scanning: Integration with container registries and security scanning tools
  • Least Privilege: Containers run with minimal required permissions
  • Security Contexts: Kubernetes security contexts for pod-level security controls
  • Secret Injection: Secure environment variable and secret management without exposing credentials
Truefoundry internally follows a policy to scan all images and artifacts and not release software with any critical known vulnerabilities that have a fix. We have an SLA to fix all critical vulnerabilities within 10 days if a fix is available.

​
Audit and Monitoring

Comprehensive audit logging tracks all platform activities:
  • Complete Activity History: All user actions, resource changes, and API calls are logged
  • Detailed Audit Trails: Track who performed what action, when, and on which resource
  • Export Capabilities: Export audit logs to your logging platform (Splunk, Datadog, etc.)
  • API Access: Programmatic access to audit logs via Truefoundry API
To read more about audit logging, please refer to our Audit Logging documentation.
Since the infrastructure is self-hosted, all your existing monitoring, alerting and security solutions can be used to monitor the Truefoundry infrastructure.

​
AI Gateway Security

​
Gateway Security Features

Truefoundry’s AI Gateway provides enterprise-grade security for LLM access:
  • Authentication: Multiple authentication methods including OIDC, JWT, and API keys
  • Authorization: Role-based access control for models and endpoints
  • Guardrails Integration: Support for content safety, PII detection, and prompt injection prevention
  • Request/Response Logging: Secure logging of API requests and responses for compliance and debugging

​
Content Safety

Integration with enterprise guardrail providers:
  • Content Filtering: Azure AI Content Safety, OpenAI Moderation, and custom guardrails
  • PII Detection and Redaction: Automatic detection and redaction of personally identifiable information
  • Prompt Injection Prevention: Protection against prompt injection attacks
  • Toxicity Detection: Real-time detection of harmful or inappropriate content
For more information on guardrails, see AI Gateway Guardrails.

​
Incident Response and Business Continuity

  • Globally Distributed: Gateway deployed across 12+ regions and multiple cloud providers including AWS, GCP and Azure.
  • Automated Failover: Automatic routing to healthy regions in case of regional downtime
  • Multi-Cloud Multi-region deployment of control-plane: Control-Plane is deployed across three regions and three cloud providers in active passive mode and has disaster recovery enabled.
  • Redundancy: High availability configurations for critical components
Truefoundry is deployed in multiple-regions and multiple-cloud providers. The control-plane is deployed in active passive mode and has disaster recovery enabled.
Status of all truefoundry hosted components can be tracked at status.truefoundry.com.Status PageYou can also subscribe to the status updates by clicking on the “Get Updates” button in the top right corner of the status page.
  • Configuration Backups: Regular backups of platform configuration and metadata
  • Disaster Recovery: Comprehensive disaster recovery procedures
  • Data Backup: Integration with your backup solutions for data plane storage
  • Recovery Procedures: Documented procedures for rapid recovery from incidents

​
Vulnerability Management, Logging and Monitoring

  • A regular vulnerability scanning program in place to ensure that security issues are identified and remediated promptly, as outlined in the Technical Vulnerability Management section of the Operations Security Policy.
  • There is auditing and logging enabled across the systems and network devices to maintain visibility into system activity. AWS CloudTrail is integrated with CloudWatch as a centralized logging and monitoring solution. This setup correlates events and generates alerts for our team whenever a potential security incident is detected.
  • Additionally, 24/7 monitoring of security alerts is maintained through this system to ensure continuous oversight and rapid response.

​
Security Reporting

If you discover a security vulnerability, please report it responsibly:
  • Security Contact: Report vulnerabilities to security@truefoundry.com
  • Responsible Disclosure: We follow responsible disclosure practices
  • Response Time: We aim to respond to security reports within 48 hours
  • Security Advisories: Published on trust.truefoundry.com
  • Release Notes: Security updates included in platform release notes
  • Notifications: Security-critical updates communicated to customers

​
Data Privacy

  • Data Minimization: Only collect and process data necessary for platform operation
  • User Data Control: Users can manage their own data and access permissions
  • Data Retention: Configurable data retention policies
  • Right to Deletion: Support for data deletion requests in compliance with privacy regulations
Truefoundry is GDPR compliant and provides:
  • Data Subject Rights: Support for data access, rectification, and deletion requests
  • Privacy by Design: Privacy considerations built into platform architecture
  • Data Processing Agreements: Standard data processing agreements available
  • Privacy Policy: Comprehensive privacy policy available on our website

​
Third-Party Security

  • Security Assessments: Regular security assessments of third-party vendors and integrations
  • Secure Integrations: All integrations use secure authentication and encryption
  • Vendor Monitoring: Continuous monitoring of vendor security practices
  • OAuth 2.0: Secure OAuth-based integrations with cloud providers
  • API Security: All API integrations use secure authentication methods
  • Credential Management: Secure storage and rotation of third-party credentials

​
Security Best Practices - Recommendations for Customers

  1. Enable SSO: Use single sign-on with your identity provider for centralized authentication
  2. Implement Least Privilege: Grant users only the minimum permissions required for their role
  3. Regular Access Reviews: Periodically review and audit user access and permissions
  4. Enable Audit Logging: Monitor audit logs regularly for suspicious activities
  5. Use Secret Management: Store sensitive credentials in secret managers, not in code
  6. Network Segmentation: Deploy Truefoundry in isolated network segments when possible
  7. Regular Updates: Keep Truefoundry components updated to the latest versions
  8. Enable Guardrails: Use content safety guardrails for AI Gateway deployments
Security is a shared responsibility. While Truefoundry provides secure infrastructure and platform features, customers are responsible for securing their applications, data, and access controls within their environment.

​
Sub-Processors

Here is the list of sub-processors:
  1. FusionAuth: Used for authentication
  2. AWS: Used for hosting, storage and security monitoring
  3. Sentry: Used for error and crash logs
  4. Posthog: Used for user event and analytics
We have signed a Data Processing Agreement (DPA) with all sub-processors. Regarding data storage, data is stored in Europe for AWS, FusionAuth, and Posthog. Sentry data is stored in the US.