Security Overview

This overview reflects our current security posture during open beta. Practices and controls will continue to mature as the Service develops.

1. Current Security Controls

We have implemented foundational security measures to protect workspace data during the beta phase:

• Encryption in Transit: All communications between clients, our backend, and third-party APIs are encrypted using TLS 1.2 or higher over HTTPS.
• Encryption at Rest: Core database volumes are encrypted at the storage level. Highly sensitive credentials, such as BYOK API keys and webhook secrets, are additionally encrypted at the application level using AES-256-GCM.
• Access Control: Tenant isolation is enforced at the application and API layer: workspace-scoped queries are constrained to the caller's workspace through a workspace-scoped data-access layer, with explicit workspace filters on the few endpoints that do not use it, and Postgres row-level security (RLS) enforces the same workspace scoping at the database layer as a second, independent boundary. Role-Based Access Control (RBAC) is enforced server-side on sensitive and privileged actions.
• Authentication: Session-based authentication uses httpOnly, Secure cookies served only over HTTPS, with SameSite protection against cross-site request forgery. User identities and login methods are managed by Descope, including magic link and social authentication. Application data remains stored in our Neon Postgres database.
• Input Validation: All ingress API boundaries strictly validate and sanitize payloads using Zod schemas to prevent injection attacks.
• Rate Limiting: Authenticated product endpoints are protected by rate limiting broadly, in addition to tighter limits on authentication, invitation, and billing paths, to defend against brute-force attempts and automated abuse.

2. AI-Specific Security

Foreman implements orchestration constraints specifically designed for autonomous agent safety:

• Request Depth Limits: Infinite loops are prevented by strict delegation depth tracking (maximum delegation depth is enforced).
• Constrained Tooling: Ordinary workspace agents use allowlisted integrations and do not receive general host-shell access. Studio coding agents are a distinct feature: they may execute development commands only on the customer-selected Foreman execution host and project workspace, under Studio's repository, secret, review, and deployment controls.
• Tool Calling Guardrails: Mutating integration actions (POST/PUT/DELETE) pass through classification gates, human approval or pre-authorized write grants, and communications-compliance checks before reaching external APIs.
• Payment and High-Risk Gates: Agents are blocked from payment execution and consequential-decision actions. Payment authorization requires a human handoff outside Foreman.
• Communications Controls: Outbound messaging paths enforce AI disclosure, suppression checks, consent evidence, opt-out language, and audit logging.
• Sanctions Screening: Signup and billing paths include jurisdiction screening and audit records, with manual review states for potential list matches.

Several of these controls are not just descriptive — they are binding commitments elsewhere in our terms: the prohibition on autonomous payment execution and consequential-decision actions is a contractual restriction in the Terms of Service (Section 6) and the Acceptable Use Policy, and our security-incident notification timeline is committed in the Data Processing Addendum (Section 8).

3. Open-Beta Security Roadmap

4. Vulnerability Reporting

If you discover a security vulnerability in the Foreman platform during the beta, we ask that you report it to us confidentially at security@foreman.company. Please do not disclose vulnerabilities publicly until we have verified and patched the issue.