Skip to content

Security & Privacy

Built to protect your data, your agents, and your people

Halo is a security product, so we hold it to the standard we help you set. Here is how the platform limits what agents can reach, keeps credentials out of their hands, and respects employee privacy.

Agent and gateway security

Controls that travel with each agent

  • Sandboxed agent runtime

    Agents run as non-root workloads with signed configuration and explicit file, network, process, credential, and model boundaries.

  • Default-deny egress

    Agent network access is denied unless a destination is approved, which limits where data can go if an agent misbehaves.

  • Brokered credentials

    Provider and app credentials stay bound at Halo gateways. Agents receive scoped capabilities and never see raw keys.

  • Quarantine and release

    Operators can quarantine an agent to stop its activity, review what happened, and release it when ready.

  • Human approvals

    Policies can require a person to approve higher-impact actions before they run, with the decision recorded.

  • Decision evidence

    Policy decisions, approvals, denials, and configuration changes are recorded and exportable as CSV or JSON.

Privacy by design

Visibility without collecting more than the job needs

Discovery is designed so security teams can see AI use while employees keep their content private.

Metadata-first discovery
Standard Halo Endpoint and Halo Browser collection focuses on application, extension, runtime, and configuration signals.
Content stays private by default
Prompts, responses, documents, uploaded files, clipboard content, and credentials are not part of standard collection.
Expanded permissions by approval
Content inspection in the browser requires expanded permissions that your organization chooses to approve.

Reviews and notices

Security reviews and documentation

Enterprise agreements can include a custom data processing agreement, a security review, and audit support. Contact our team to start a review, and see our notices for details on data handling.

Start a security review

Put a halo around the AI your business runs.

See discovery, runtime policy, approvals, and evidence working together on your own use cases.