Security

Infrastructure

• Halo AI is hosted on leading cloud providers in the United States. Our infrastructure providers maintain SOC 2 Type II certification and operate physically secure, redundant data centers.
• Databases are backed up automatically every day, and deployments roll out without downtime and can be rolled back.
• Production access is limited to a small number of named engineers and is granted on a least-privilege basis.

Encryption

• All data is encrypted in transit using TLS 1.2 or higher, and our domains enforce HTTPS with HTTP Strict Transport Security.
• All data is encrypted at rest using AES-256, including databases, file storage and backups.
• Credentials for the integrations you connect are additionally encrypted at the application layer before they are stored, with keys kept in a managed secret store and rotated on a schedule. They are never written to logs or returned by our API.

Tenant isolation

Halo AI is a multi-tenant platform, and keeping each customer’s data separate is our first design rule. Every record is tied to a single workspace, and that scoping is enforced at the database layer as well as in our application code, so a request can only ever read or write the data of the workspace it is authenticated for. Knowledge retrieval, background jobs and file storage follow the same rule. These boundaries are covered by automated tests, and we treat any cross-tenant defect as a top-severity incident.

Access control and authentication

• Dashboard users sign in with email and password, with minimum password strength requirements, or with Google. Multi-factor authentication is available to every user, and single sign-on is available on Enterprise plans.
• Role-based permissions control what each member of your team can see and change, and administrative actions are recorded in an audit log.
• Sessions use short-lived tokens, and sign-in attempts are rate limited to protect against credential stuffing and brute-force attacks.
• API keys and MCP connections are scoped to a single workspace, can be rotated at any time, and are shown only once when created.
• Identities you pass to the widget can be cryptographically signed by your servers, so that one of your users cannot impersonate another.

AI safeguards

• Your data is used to make your own agents better and for nothing else. We do not train general-purpose AI models on customer data, and our large language model providers do not use data sent through their APIs to train their models.
• Agents only retrieve knowledge that belongs to your workspace, and customer-specific records are shown only to the customer they belong to.
• Content retrieved from documents, web pages and conversations is isolated from the agent’s instructions to limit prompt injection, and responses are checked for personal details that do not belong in the conversation before they are sent.
• You control what your agents can do: which tools they may call, when they escalate to a human, and which actions require approval.

Application security

• Every change is peer reviewed and passes automated tests, type checks and lint checks before it is deployed.
• Dependencies are kept current and reviewed for known vulnerabilities, and security findings are triaged and fixed according to severity.
• The platform applies a content security policy, strict cross-origin rules, signed and replay-protected webhooks, input validation, per-route rate limiting and a web application firewall.
• The embeddable SDK is served with subresource-integrity hashes, so you can pin the exact code that runs on your site.
• Secrets are stored only in managed secret stores and are never committed to source control.

Monitoring and logging

• Application errors, performance and security-relevant events are monitored continuously, and the team is alerted to anomalies.
• Logs are scrubbed of secrets and personal data before they leave the platform and are retained for a limited period for security investigation.
• Every AI conversation records which knowledge was used to answer, so that retrievals can be audited after the fact.

Incident response

We maintain a documented incident response process covering triage, containment, credential rotation, forensic review and post-incident analysis. If an incident affects your data, we notify you without undue delay and in any event within 72 hours of becoming aware of it, with the information you need for your own notifications, as set out in our Data Processing Addendum.

Privacy and compliance

• Our Data Processing Addendum applies automatically to every customer. It includes the EU Standard Contractual Clauses, the UK Addendum and Swiss adaptations for international transfers, and service-provider terms under the CCPA and other US state privacy laws.
• Every third party that processes customer data is listed on our Sub-processors page, and we give 30 days’ notice before adding one.
• You can export your data at any time through the dashboard, the API and the MCP server, delete individual records from within the product, and request deletion of your whole workspace. Deletions cascade to knowledge derived from the deleted records.
• We complete security questionnaires and share our current security documentation with customers and prospects on request.

Your part

Security is shared. We recommend that you enable multi-factor authentication for every member of your team, grant the minimum permissions each person needs, keep API keys out of client-side code, sign the identities you pass to the widget, review your agents’ tools and escalation rules, and tell your users how AI is used in your product.

Responsible disclosure

If you believe you have found a security vulnerability in Halo AI, please tell us before you tell anyone else. Email support@haloagents.ai with the subject “Security vulnerability report”, including the steps needed to reproduce the issue. We acknowledge reports promptly, keep you informed while we investigate, and credit researchers who wish to be credited once a fix has shipped. We do not currently run a paid bug bounty program.

We will not pursue legal action against researchers who act in good faith: test only against accounts you own, do not access, modify or delete data that is not yours, do not degrade the service for others, and give us reasonable time to fix an issue before disclosing it publicly.

Contact

Security questions, questionnaires and documentation requests can be sent to our team: