A breakdown of the September 2026 industry brief that argues Canada already has the cyber rules it needs — and is losing ground on enforcing them against machine identities.
Canada’s national technology industry association, TECHNATION, published the second edition of its Cyber Intelligence Quarterly Brief on 2 September 2026, and the issue is built around one argument: Canadian organizations are handing operational authority to software faster than they are building the governance to supervise it. The brief was compiled from written contributions by members of the association’s Cybersecurity Task Force and released ahead of Cyber Security Awareness Month in October.
Its central claim is that machine and agentic AI identities — the service accounts, API keys, tokens and agent credentials that let automation act on a company’s behalf — are now multiplying faster than the oversight structures meant to control them. The document is aimed at boards and executive teams rather than security operations staff.
Task force chair Kevin Dawson, president and chief executive of ISA Cybersecurity, writes that three months after his previous brief flagged the pace problem, the group went deeper into “the finding that worried me most: the identities we are handing to machines, and how quickly we are handing them out.” His stated hope is that the contributions prove not merely thought-provoking but “action-provoking.”
The 43 Percent Figure Anchoring This Quarter’s Argument
The most concrete data point in the brief comes from Jatinder Singh Mann, founder and chief executive of Cetark Corp and Netonelabs Global Operations. Mann cites a Verizon 2026 report finding that organizations expanding AI-driven identities faced a 43 percent breach rate, compared with 11 percent for organizations that did not.
He pairs that figure with a scale estimate: machine identities now outnumber human users in the average enterprise by roughly one hundred to one. Most of those credentials, he writes, were issued without a named owner, an expiry date or any retirement policy — and that gap is where breaches now originate.
For readers outside the security profession, the practical translation is simple. Every automation, integration and AI agent an organization deploys needs credentials to act on its behalf, and those credentials rarely get decommissioned. The population grows quietly, and each unclaimed credential is a live door.
Who Owns the Agent, What Can It Reach, Who Can Switch It Off
The brief’s first question asks what effective governance of the non-human identity estate looks like once agentic AI moves from pilot to production. Every contributor started in the same place: build a working inventory.
Mike Carzim, senior manager of governance within global information security and privacy at OpenText, argues that AI agents, service accounts, API keys, machine identities and automation credentials should be treated as governed digital identities rather than technical artifacts. His prescribed sequence is discovery first — a central registry of agents, mapped privileges and dependencies, and identification of orphaned identities — followed by risk-based controls, continuous monitoring, audit trails and human oversight of high-risk actions.
Mann sets out five disciplines applied to identities that never sign in:
- Named human owner: every non-human identity is attached to a person, and credentials nobody claims are revoked.
- Task-scoped least privilege: permissions attached to a specific job, not a broad role.
- Short-lived credentials by default: no static secrets embedded in code.
- Human approval gates: a person signs off before an agent takes a high-impact action.
- Continuous monitoring for behaviour drift: with revocation built into the process.
Andrew Buckles, executive vice-president at ISA Cybersecurity, adds a layer above the inventory: a control objective framework that converts legislative, regulatory, business and sovereign-AI obligations into enforceable requirements. He notes the framework must cover controls that did not exist in traditional IT governance — model inventorying and decommissioning, model testing, agentic system engineering standards, agentic detection and response, and governance over which skills and tools agents may use.
Buckles is blunt that documentation alone proves nothing. Organizations need recurring AI risk assessments and AI red teaming to confirm controls actually hold, with findings feeding a remediation loop. The task force’s recurring test, as the brief summarises it, is three questions long: who owns this agent, what can it reach, and who can switch it off?
Shadow AI Framed as Unmet Demand, Not a Discipline Problem
The second question addresses a point raised in the task force’s first bulletin: many organizations cannot say where AI is already in use, what it is accessing, or where data is moving. Notably, every response set prohibition aside.
Ben Lyons, senior director of policy and public affairs at Darktrace, describes shadow AI as a sign of unmet business demand and warns that blanket bans push activity onto personal accounts and devices, making tools, data access and data destinations harder to see. His alternative is approved tools, clear guidance, a simple route for new use cases, and continuous visibility across SaaS, cloud, endpoints and internal development environments, with behavioural monitoring flagging unusual data flows.
Mann frames the same dynamic as the shadow IT story repeating itself, and proposes a three-step sequence: see it, using network telemetry, DNS logs, browser controls and cloud access brokers; sanction it, with enterprise routes carrying vetted terms on training data, retention and jurisdiction; and sustain it, because the picture changes weekly rather than annually. His measure of success is unambiguous — if the sanctioned route is not faster than the shadow one, people will keep routing around it.
Carzim adds an information-classification dimension, noting that many banks, healthcare providers and other regulated organizations restrict unapproved public AI services while maintaining a catalogue of enterprise-approved solutions backed by contractual, privacy and security agreements. Data classified at the most sensitive tiers, he argues, should be blocked from AI processing even within approved systems.
Buckles contributes a distinction worth noting for any executive auditing their own environment: shadow AI arrives deliberately, when someone reaches for a tool, and by default, when a vendor switches AI features on inside software the organization already owns. Each requires its own response. He also pushes back on the idea that governance and speed trade off against each other, arguing that a poor foundation eventually has to be rebuilt and that a single serious incident can freeze an AI programme for months.
The Questions the Task Force Wants on Board Agendas
The third section builds on the June brief’s warning that the leadership awareness gap is underappreciated and that AI should not be treated as just another technology cycle. The task force’s position is that decisions once requiring a human signature are now being made by models and agents, which makes AI a change in how an organization runs rather than a project to delegate and revisit at budget time.
Mann lists five questions he believes belong on a standing board agenda over the next 24 months:
- Production inventory: do we know every AI system in production, who owns it, and what data it touches?
- Delegated decisions: which decisions have been handed to a model, and who is accountable when the model is wrong?
- Non-human identity count: what is our current inventory, and how many carry more privilege than they need?
- Data destinations: where is data going through AI vendors, and under what training-data and jurisdictional terms?
- Incident readiness: if a serious AI-related incident happened tomorrow, is the playbook, disclosure protocol and regulator relationship in place?
Buckles offers a parallel set, adding questions on model testing and decommissioning processes, whether a compromise would be detected or merely discovered, whether data leaves Canada or moves to third-party models on reviewed terms, when agentic systems were last risk-assessed or red teamed, and how the organization tracks against evolving compliance requirements. He also raises what the brief calls the “Fable 5 experience” as a prompt to ask whether an organization is building reliance on services outside Canada that could be restricted or withdrawn without notice.
The brief’s own editorial note on the 24-month framing is pointed: don’t wait, because the answers matter too much to defer.
Canada’s Regulatory Floor: Enforcement Rather Than New Rules
One of the clearer policy signals in the document is what it does not call for. Rather than urging new AI legislation, contributors argue that Canada’s existing instruments already apply.
| Instrument | As described in the brief |
| Canadian Centre for Cyber Security agentic AI advisory (May 2026) | Co-authored as a Five Eyes advisory on careful adoption of agentic AI; described as the current baseline, aligned with the five identity disciplines |
| PIPEDA | Cited as applicable federal privacy legislation |
| Quebec’s Law 25 | Cited as applicable provincial privacy legislation |
| OSFI Guideline B-13 | Cited as applicable to federally regulated financial institutions |
As Mann puts it, nothing new needs to be invented — only enforced against a landscape moving faster than any of those instruments were written for. It is a position with real weight in Canada’s ongoing AI policy debate, coming from an industry association whose membership spans much of the country’s technology sector.
Why the Timing Matters for Canadian Organizations
Publishing a month before Cyber Security Awareness Month gives the brief an obvious hook, but the more consequential argument is about the cost of delay. In his closing reflections, Dawson writes that capability is arriving faster than governance cycles were built to absorb, and that every month of waiting adds more agents, more credentials and more connections somebody will eventually have to account for.
Whatever the cleanup costs today, he argues, it will cost considerably more a year from now. The encouraging counterpoint he offers is that most of the recommended work is governance rather than technology spending: deciding who owns an agent, defining what it is allowed to do, and withdrawing access nobody can account for.
That framing extends the reach of the advice to smaller organizations. Dawson notes that even those sitting below the “cyber poverty line” the task force discussed earlier in the year can still do the work. Mann’s suggested first pass is 90 days: discover what exists, classify credentials by blast radius, assign a human owner to each, and revoke what nobody claims.
Background: Why Machine Identity Has Become a Governance Topic
The vocabulary in the brief will be unfamiliar to many general readers, so a short glossary helps. A non-human identity is any credential used by software rather than a person — a service account, an API key, a token or a certificate. An agentic AI system is one that can take multi-step actions toward a goal rather than only producing text or predictions.
Traditional identity governance was designed around employees, with joiners, movers and leavers triggering access reviews. Software credentials generally have no equivalent lifecycle event, which is the structural reason contributors keep returning to ownership and expiry. This is broad professional context rather than a claim from the brief itself.
Red teaming, referenced by Buckles, means deliberately attempting to defeat a system’s controls to see whether they hold in practice. Applied to AI, it typically covers attempts to manipulate model behaviour or push an agent beyond its intended scope.
What Happens Next
The Cyber Intelligence Quarterly Brief is a recurring publication, so a further edition is expected in the next quarter and will likely revisit whether the identity inventory problem has moved. Dawson closes by encouraging readers to put the third section’s questions to their own security teams immediately.
The nearer-term marker is October’s Cyber Security Awareness Month, when the themes in the brief are likely to be echoed across Canadian public and private sector campaigns. Whether the enforcement-over-new-rules position gains traction in federal AI policy discussions is Not specified in the document itself.
Editorial Note and Transparency
This article is an independent report on a publicly released industry brief. All figures, quotations and role titles are attributed to the brief and its named contributors; this publication has not independently verified the underlying data, including the Verizon 2026 breach-rate finding.
Nothing here constitutes security, legal or compliance advice. Organizations assessing their own exposure should consult qualified professionals and the applicable regulators. This site is not affiliated with TECHNATION, any government body, or any organization named in the brief.
Frequently Asked Questions
What is the TECHNATION Cyber Intelligence Quarterly Brief?
It is a recurring publication from TECHNATION, Canada’s national technology industry association, compiled from written contributions by members of its Cybersecurity Task Force. The second edition was released on 2 September 2026.
What is the brief’s main warning?
That machine and agentic AI identities are being created faster than the governance structures meant to supervise them, leaving credentials without named owners, expiry dates or retirement policies.
Where does the 43 percent figure come from?
Jatinder Singh Mann cites a Verizon 2026 report finding that organizations expanding AI-driven identities faced a 43 percent breach rate, compared with 11 percent for organizations that did not.
What is a non-human identity?
It is a credential used by software rather than a person, such as a service account, API key, token or agent credential. The brief estimates machine identities outnumber human users in the average enterprise by roughly one hundred to one.
Does the brief recommend banning shadow AI?
No. Every contributor set prohibition aside. Ben Lyons of Darktrace argues that blanket bans push activity onto personal accounts and devices, reducing visibility, and recommends approved tools, clear guidance and continuous monitoring instead.
Is TECHNATION calling for new Canadian AI legislation?
Not in this brief. Contributors argue that existing instruments — the Canadian Centre for Cyber Security’s May 2026 agentic AI advisory, PIPEDA, Quebec’s Law 25 and OSFI Guideline B-13 — already apply and need enforcement.
What are the three questions the task force uses as a test?
Who owns this agent, what can it reach, and who can switch it off?
What is the suggested starting point for organizations?
Mann proposes a 90-day first pass: discover what exists, classify credentials by blast radius, assign a human owner to each, and revoke what nobody claims.
Does the advice apply to smaller organizations?
Dawson argues it does, noting that most of the recommended work is governance rather than technology spending, and that even organizations below the “cyber poverty line” can carry it out.
When is the next edition expected?
The brief is quarterly, so a further edition is expected in the next quarter. An exact publication date is Not specified.
