Ottawa is asking five questions about AI transparency. Can your organization answer them?

By Alexander Leslie, Chief Information Officer · Published 1 August 2026

On 23 July 2026, the Government of Canada opened a public consultation on AI transparency. Artificial Intelligence and Digital Innovation Minister Evan Solomon announced it with the statement that AI adoption moves at the speed of trust. The consultation is run by Innovation, Science and Economic Development Canada and has now closed. It asked Canadians how the country should approach five areas. The first two are detecting and identifying AI generated content, and notifying people when they interact with an AI system. The others are making accessible information available about AI capabilities and limitations, tracking serious AI related incidents, and monitoring AI agent activity.

Read as an internal audit rather than a policy document, the five areas form a readiness checklist that applies today. The survey evidence below suggests many organizations would struggle to pass it.

These are the questions a board, an auditor, a regulator or a skeptical customer already asks in some form. The consultation puts them in one place and signals that formal expectations are coming.

VentureBeat Research, published 24 July 2026, surveyed 573 enterprise respondents at organizations with 100 or more employees. Roughly two thirds already allow AI agents to push changes into production on automated evaluations with no human review. Only 5 percent fully trust the systems doing that reviewing. About 47 percent of enterprises monitor and log agent activity at all. Half have shipped an agent that failed in a customer facing system. It had passed internal evaluations.

The identity layer underneath is just as thin. Okta's Enterprise AI Index, published 16 July 2026, analyzed anonymized sign in data from more than 20,000 organizations. It found AI agents overwhelmingly authenticate through legacy methods: service accounts, static keys and shared human logins. When an agent signs in as a person, the audit trail stops distinguishing between the two, and the monitoring question in Ottawa's consultation becomes unanswerable by construction.

The consequences are already arriving as incidents rather than hypotheticals. DigiCert's survey of 1,001 information technology (IT) and cybersecurity leaders, published 7 July 2026, found 78 percent encountered AI related security issues within the past six months. Half of enterprises experienced an actual breach or disruption tied to unauthorized or misconfigured AI agents. The same survey found 90 percent of leadership teams discuss AI governance while only about half have a formal governance program. The conversation is happening. The capacity to answer is not being built at the same rate.

Transparency answers an internal audience as well as an external one. The AI Insights Report from TD Bank Group (TD) was conducted by Ipsos among 2,501 Canadian adults and released 23 July 2026. It found 78 percent of Canadians believe workplace AI adoption is inevitable. In the same survey, 75 percent grade their own AI skills at a C or lower. Only 37 percent say they have received adequate AI training from their employer.

People expect these systems in their working lives and do not yet feel equipped to understand them. An organization that can plainly state what its systems do, when people are interacting with them, and how problems get caught is answering its own employees. That answer serves any future regulator too.

1. Inventory first. List every deployed AI system and agent, what it touches, and who owns it.
2. Identity second. Give every agent its own scoped identity so actions are attributable to exactly one actor. Without it, monitoring cannot tell you which agent did what.
3. Disclosure third. Draft the plain language statement of what each customer facing system is and is not, and decide where a person encounters it.
4. Incident capture fourth. Name the place where AI related failures get recorded and reviewed, even if it starts as a simple register.
5. Then keep the record current. Rules shaped by this consultation will ask for evidence an organization already holds, not for a new project.

Every one of these steps makes the next deployment easier to approve, not harder. Attribution shortens reviews. Disclosure builds the trust that expands what users will accept. Incident records turn surprises into institutional learning. Transparency built early makes adoption easier to sustain.

Why does the minister's phrase hold up?

Because an approval rests on trust. An organization that can already answer the transparency questions does not need a rule to force the work, and its reviewers have evidence to approve against. Being able to answer these questions is what keeps the pace of adoption sustainable.

White Glove Labs (WGL) builds this answering capacity into the systems it designs, from agent identity and monitoring through disclosure and incident practice. The controls sit in our control model, and the acceptance evidence in how to evaluate a knowledge agent. If your organization wants to be able to answer all five of Ottawa's questions with evidence, the inventory and the agent identities are where it starts.