Security

Security first

Trust is the foundation of a compute marketplace. We start with verified organizations and build security into every layer.

What we protect

Our security model protects four categories of assets:

  • Customer data: prompts, inputs, model weights, outputs
  • Provider hardware: GPU nodes, host systems, network
  • Platform data: job metadata, billing records, logs
  • Credentials: API keys, authentication tokens, secrets

Threat actors

Who might try to attack the system and why.

Malicious Customer

A customer submitting harmful workloads to attack provider infrastructure, steal data, or abuse resources.

Malicious Provider

A provider attempting to access customer data, tamper with outputs, or exfiltrate sensitive information.

External Attacker

Someone targeting the platform, API, or communication channels to disrupt service or steal data.

Insider Risk

Platform operators with elevated access who could abuse their position.

Attack surfaces and mitigations

Surface Risk Mitigation
Job Payloads Malicious code in submitted workloads Container isolation, resource limits, signed jobs, restricted system calls
Runtime Environment Container escape, privilege escalation Hardened containers, no root access, seccomp profiles, regular updates
Network Data exfiltration, command and control Restricted outbound access, allowlisted endpoints, encrypted transit
Secrets API key theft, credential exposure Short-lived tokens, encrypted storage, minimal secret distribution
Logs Sensitive data in logs, log tampering Log sanitization, append-only storage, access controls
Billing Fraudulent usage, billing manipulation Cryptographic job receipts, independent usage tracking, reconciliation

Security principles

The guidelines that shape our decisions.

Verified Organizations First

We start with known, vetted organizations. This reduces risk from anonymous or pseudonymous participants.

Workload Isolation

Every job runs in its own isolated container. No shared state, no access to host, no persistence.

Signed Jobs

All workloads are cryptographically signed. Providers verify signatures before execution.

Least Privilege

Components have minimal permissions. Jobs cannot access more than they need.

Auditable Logs

Complete audit trail for every job. Append-only, tamper-evident, available to both parties.

Abuse Prevention

Rate limits, anomaly detection, and manual review for suspicious activity.

Security roadmap

What is implemented now and what comes next.

MVP (Now)

  • Container-based isolation with resource limits
  • Basic network restrictions
  • Signed job manifests
  • Append-only audit logs
  • Verified organization onboarding

Near-term

  • Enhanced runtime sandboxing (gVisor)
  • Hardware attestation for provider nodes
  • Automated anomaly detection
  • Third-party security audit

Future

  • Confidential computing (TEEs)
  • Formal verification of critical paths
  • SOC 2 compliance
  • Bug bounty program

Operational security

Beyond technical controls, we maintain operational security practices:

  • Minimal team access to production systems
  • Access logging and regular review
  • Incident response procedures
  • Regular security reviews of code changes
  • Dependency scanning and updates

We are a small team and honest about our stage. Formal compliance certifications come later. For now, we focus on doing the right things and being transparent about our approach.

Responsible disclosure

If you discover a security vulnerability, we want to hear from you. Please report it responsibly:

  • Email: security@coranorlabs.com
  • Include details to help us understand and reproduce the issue
  • Give us reasonable time to address the issue before public disclosure

We appreciate security researchers who help us improve. While we do not currently have a formal bug bounty program, we are grateful for responsible reports and will acknowledge contributors.

Security contact: security@coranorlabs.com