Security & Encryption

Workhog is built with defense-in-depth security. Your data is encrypted at rest and in transit, isolated per user, and processed by a sandboxed AI agent that cannot escape its boundaries.

Trust Center

This page is a summary of Workhog's security posture. For full architecture diagrams, compliance details, and third-party audit results, visit the Trust Center.

Overview

Workhog protects your data with multiple layers of security, from network encryption to per-file key management:

AES-256 at rest

All files encrypted with AES-256 in S3 server-side encryption

TLS 1.2+ in transit

All connections use TLS 1.2 or higher, including WebSocket

Per-user isolation

Each user scoped to their own workspace with path enforcement

AI sandbox

OS-level isolation prevents the AI agent from accessing anything outside its scope

AI Sandbox

The AI agent that processes your requests runs inside an OS-level sandbox. This ensures that even if the model were to attempt unauthorized operations, the operating system blocks them before they can execute.

Sandbox Layers

Filesystem write isolation
Seatbelt / Bubblewrap

The agent can only write to a temporary workspace directory. All other filesystem paths are read-only or completely blocked.

Network isolation
OS-level

Outbound network access is restricted by default. The agent cannot make arbitrary HTTP requests or open connections to external services.

Process tree isolation
Namespace

The agent process runs in an isolated process tree. It cannot inspect, signal, or attach to other processes on the host.

Platform-specific enforcement

On macOS, sandboxing is enforced via Seatbelt profiles. On Linux, Workhog uses Bubblewrap (bwrap) for namespace-based isolation. Both provide equivalent security guarantees. See the Trust Center for full diagrams of the sandbox architecture.

KMS Encryption

Beyond the default AES-256 encryption at rest, you can add AWS Key Management Service (KMS) encryption to any file you choose. You decide which files get the extra layer. The key is Workhog's, held in AWS KMS in Workhog's AWS account, and every use of it is recorded in AWS CloudTrail. For how to enable KMS on individual files, see the File Management guide.

Features

  • Per-file encryption — Each file you encrypt gets its own data key, protected by Workhog's KMS key, so you can add a layer to sensitive documents and leave the rest at standard encryption.
  • Audit trail — Every use of the key is recorded in AWS CloudTrail in Workhog's AWS account.
  • Automatic key rotation — KMS keys are automatically rotated on an annual basis. Previously encrypted data remains accessible with old key versions.

How to Enable KMS Encryption

To add KMS encryption to a file:

  1. Right-click the file in the sidebar file browser.
  2. Select Encrypt with KMS from the context menu.
  3. The file is re-uploaded with KMS encryption applied. A icon appears next to the file name to indicate KMS protection.

Data Isolation

Every user's data is scoped to their own workspace directory at workspaces/{userId}/. This scoping is enforced at multiple layers:

  • Path enforcement (securePath()) — All S3 tool operations pass through a securePath() function that validates and normalizes file paths. Any attempt to traverse outside the user's workspace (e.g., using ../) is rejected before the operation reaches AWS.
  • IAM scoping — The IAM role used by the agent has a policy condition that restricts S3 access to the specific user's prefix, providing a second layer of defense even if path validation were bypassed.
  • Workspace isolation — Each agent session operates in its own temporary workspace directory on the host filesystem. Sessions cannot read each other's temporary files.

Your Data, On Request

Your files belong to you and stay in your own workspace. You can ask for them at any time by writing to privacy@workhog.ai:

  • A copy — We send you a copy of everything in your workspace within 30 days.
  • Deletion — We delete your conversations, your files or your whole account, as you ask, within 30 days. You can also delete your whole account yourself at any time in Settings, under Delete account.

Before we act on a request, we confirm it comes from your account's email address.

No Training on Your Data

Workhog uses Claude as its AI backbone. Your data is never used to train Anthropic's models:

  • Ephemeral processing — File contents are loaded into the agent's context window for the duration of the conversation only. Once the session ends, the context is discarded.
  • No retention by Anthropic — Under Workhog's API agreement with Anthropic, conversation data is not retained or used for model training.
  • You control deletion — Delete any file from your workspace at any time. Deleted files are permanently removed from cloud storage.

Anthropic's data policy

Anthropic's API terms explicitly state that data sent through the API is not used for training. This applies to all Workhog conversations and file processing.

Next Steps

Security & Encryption - Workhog Help