Security

Security at TBW

TBW designs and builds AI solutions, including private AI, agents and enterprise integrations. Security requirements are defined for each engagement. We treat identity, data, tools and human oversight as one system, because an AI feature is only as secure as the permissions, data paths and approvals around it.

This page explains our approach and the public frameworks we use as references. Specific controls are agreed for each engagement, so nothing here is a guarantee that a particular control is present in every environment. Credentials and attestations are listed in the Trust Center.

How we organize security work

We use the NIST Cybersecurity Framework 2.0 as a reference structure. It has six functions, and we plan engagements against all of them rather than only prevention.

Function What it means in our engagements
Govern Named owners, written security requirements, and documented risk decisions for each engagement.
Identify An inventory of data, users, systems, AI models, tools and third-party services before design starts.
Protect Least-privilege access, separate identities for people, services and agents, protected secrets, and data minimization.
Detect Logging and monitoring of access, agent actions and tool calls, sized to the environment.
Respond Defined incident roles, escalation paths, and customer and regulatory notification.
Recover Backup, restore and rollback planning, tested against the recovery goals agreed with the customer.

Securing AI systems

AI systems add risks that conventional controls do not cover on their own. We use these references when we design, review and test them:

Our design principles:

  • Authorization lives outside the model. A prompt is not a security boundary. Access decisions are enforced by systems that do not depend on the model behaving correctly.
  • Least privilege for agents and tools. Scoped, short-lived credentials, and separate identities for people, services and agents.
  • Human approval for consequential actions. Writes, deletions, payments and external messages pass through an approval step.
  • Untrusted by default. Retrieved documents, web pages, files and tool results are treated as untrusted input that may carry hostile instructions.
  • Controlled tools and connectors. Allowlisted tools and MCP servers, arguments validated server-side, and third-party models, connectors and dependencies reviewed before use.
  • Secrets stay out of prompts, logs and client code.
  • Output handling. Model output is validated before it reaches downstream systems.
  • Protected knowledge sources. For retrieval systems (RAG), we track document provenance, control what is ingested, and apply access controls at the document level.
  • Limits that contain failure. Step budgets, timeouts, and rate and cost caps to limit runaway loops and abuse.
  • Tested, not assumed. Adversarial and failure-case testing before release and after significant changes.

Data protection and privacy

  • Data boundaries. We map where information is stored, processed, retrieved and shared, including model providers and other third parties, and agree those boundaries with the customer.
  • Minimization and retention. Retention, redaction and data loss prevention (DLP) are set for each engagement based on data sensitivity.
  • Private and local AI options. Where data must stay inside a defined boundary, we design deployments that keep it there.
  • Privacy law. We handle personal information in line with applicable Canadian privacy law, including PIPEDA.

Incident response

We maintain an incident handling process covering triage, containment, communication and review.

Where an incident involves personal information under TBW’s control and creates a real risk of significant harm to individuals, we report to the Office of the Privacy Commissioner of Canada and notify affected individuals as required by law. We keep a record of security breaches involving personal information, whether or not they reach that threshold. For customer environments, notification terms follow the agreement for that engagement.

Assurance and certifications

The Trust Center lists the credentials and attestations TBW holds. Guidance documents and frameworks named on this page are references we work with. They are not certifications.

Report a security vulnerability

If you believe you have found a vulnerability affecting TBW.com or a TBW-managed service, please tell us. Read the Responsible Disclosure guidance, then email [email protected] with the subject line Security Vulnerability.

Please include:
– The affected URL, service or system.
– A description of the issue and the steps to reproduce it.
– The potential impact, as you understand it.
– How we can reach you.

While you test, please:
– Avoid accessing, changing or keeping data that is not yours. Stop and tell us if you encounter personal information.
– Avoid actions that degrade service, such as denial-of-service testing.
– Avoid social engineering, physical attacks and testing of third-party systems.
– Give us a reasonable time to investigate and fix the issue before you share details publicly.

Please do not email passwords or other people’s personal data. If you need to share sensitive details, tell us in your first message and we will agree on a secure way to exchange them.

Related pages

Last reviewed: 30 September 2026

Restricted assurance reports are not available for public download.

Back to top button