AI Agents Are Creating an Identity Crisis in Your IT Systems

Most small and mid-sized businesses we work with are already using some form of AI automation. Maybe it is a chatbot that handles customer inquiries, a tool that processes invoices, or a Microsoft 365 Copilot workflow that summarizes emails and schedules meetings. These are useful tools. But they come with a security problem that almost nobody is talking about yet.

According to TechCrunch, a startup called Oak just came out of stealth mode with $60 million USD in funding. Its entire focus is on one problem: AI agents are being granted access to business systems, and nobody is properly managing or tracking that access.

What Is an AI Agent, and Why Does Its Identity Matter?

A regular user in your business has an account. That account has a username, a password, and ideally multi-factor authentication. When something goes wrong, you can look at the logs and see exactly what that person did and when. Access can be revoked in seconds.

An AI agent does not work the same way. When a tool like a Copilot workflow, a Zapier automation, or a third-party AI assistant connects to your Microsoft 365, your CRM, or your file storage, it usually authenticates using a service account or an API key. These credentials rarely have expiry dates. They often have broader permissions than they need. And almost nobody is auditing them.

Oak's founders told TechCrunch that AI agents are making this problem significantly worse because the number of non-human identities inside a typical business system is growing faster than anyone can track manually. One compromised API key or over-permissioned service account is enough to give an attacker persistent access to your data.

This Is Not a Future Problem

We want to be direct with you: this is happening right now inside businesses your size.

If your team has connected any AI tool to your Microsoft 365 tenant, your Google Workspace, your accounting software, or your cloud storage, there is almost certainly at least one non-human identity in your environment that nobody is actively monitoring. We see this consistently when we do IT audits for new clients in Toronto and across Ontario.

The typical scenario looks like this. Someone on your team signs up for an AI productivity tool and connects it to your Microsoft 365 account. The tool gets granted access to read emails, write to SharePoint, or pull data from Teams. A year later, that employee has left. The tool subscription may have lapsed. But the access credential is still active, sitting in your Azure Active Directory, connected to nothing visible, audited by no one.

What Anthropic and Blackstone Are Seeing

This is not just a startup concern. TechCrunch also reported in mid-July 2026 that Anthropic and Blackstone are both betting that the next major wave of AI business value will come from implementation, meaning the actual integration of AI into business workflows, not just from building the models themselves. That means more AI agents, more integrations, and more non-human identities inside your systems. The identity problem Oak is trying to solve will get bigger before it gets smaller.

What We Recommend for Canadian SMBs Right Now

You do not need to wait for Oak's product or any other specialized tool to start managing this risk. Here is what our team recommends for businesses running Microsoft 365 or Google Workspace.

• Audit your connected apps. In Microsoft 365, go to the Azure Active Directory admin centre and look under Enterprise Applications. Every entry in that list has access to something in your environment. Many of them were approved by individual staff without IT oversight. Pull that list and review it.
• Revoke what you do not recognize or no longer use. If an app has not been used in 90 days, revoke its access. You can always re-authorize it later if someone complains. The risk of leaving stale credentials active is far higher than the inconvenience of re-connecting a tool.
• Apply least-privilege permissions. When you do connect an AI tool or any third-party service, give it only the permissions it actually needs. A tool that summarizes documents does not need write access to your entire SharePoint environment.
• Set a recurring review date. We recommend a quarterly review of all connected applications and service accounts. Put it in the calendar now for October 2026.
• Ask your IT provider about non-human identity monitoring. If you are using a managed IT provider and they have never raised this topic with you, that is worth a conversation.

The Bottom Line

The fact that a company raised $60 million specifically to solve the AI agent identity problem tells you how serious and widespread it is. Oak's emergence is a signal, not just a product announcement. The businesses that get ahead of this now, by auditing their connected apps and tightening permissions, will be in a much stronger position as AI tooling continues to expand inside Canadian workplaces.

If you want our team to run a connected application audit on your Microsoft 365 or Google Workspace environment, reach out to us. It is one of the fastest ways to close an access gap that most SMBs do not know they have.