Security

Understand the controls and their boundaries.

A useful security review describes how a control works, what it depends on, and how its behaviour has been tested.

Local policy checks

Configured hooks can deny selected actions or require approval before execution. Other hooks record outcomes after an action. Protection depends on the validated settings, permissions, sandbox and endpoint configuration.

Evidence with defined handling

Records contain decisions, control references, hashes and redacted summaries. They can also contain endpoint, session, repository and working-directory metadata. Agree access, forwarding and retention before collecting operational evidence.

Integrity and retention

Hash links support checks for changes within a retained log. Local chaining alone cannot prove completeness or protect against a rewritten chain. Trusted retained copies and central write-once storage provide stronger integrity.

Layered protection

Secret detection uses patterns and has limits. Managed endpoint controls, network policy, identity, code review and human approval remain part of the security design.

Separate Toren evidence from agent data flows

The evidence design uses hashes and redaction to reduce recorded content. This does not mean that code or prompts never leave an endpoint: Claude Code’s connections to model providers and other services depend on your configuration and agreements.

Review your deployment

During pilot scoping, confirm the supported configuration, control ownership, evidence handling and validation criteria. Request a walkthrough of the evidence format and test approach before deciding on a rollout.

Report a security concern

Contact Toren about security with a short description and a way to reach you. Please avoid including credentials, personal data or sensitive exploit details in the initial enquiry.

Discuss security requirements