AI Honesty in the Organization
I think one of the more important conversations we need to have about AI Governance is also one of the simplest: Are we being honest about how and why we ...
Organizations need to move past the question of whether employees should be using AI and recognize that many already are. They are using it to write emails, summarize documents, analyze information, and help them get through their work. Some are using personal accounts on personal devices, but still doing work. Others are using features built into software the organization already owns.
Leadership may not have approved it or even know about it. That does not mean it is not happening nor that you are not liable for the activity.
From a security perspective, this should be a reason to provide a tool the organization can manage. We need organizational accounts, administrative controls, and visibility into how the service is being used. We need to be able to investigate when someone reports that confidential information was uploaded or an account was misused. Leaving employees to choose their own tools makes those responsibilities much harder.
If an employee puts sensitive information into a personal AI account, what can the organization actually determine afterward? Can we establish what was submitted, which files were uploaded, or whether the information was shared? Can we restrict the account or preserve relevant records? We may be left depending on what the employee remembers and what they are willing or able to show us.
Speaking from experience, that is a poor position from which to investigate an incident.
Providing an enterprise AI service gives the organization an opportunity to establish control over an activity that is already occurring. Employees can authenticate through organizational accounts, access can be assigned and removed, and administrators can configure the protections the service supports. Depending on the platform, the organization may also have access to audit records, retention settings, integration controls, and data protection capabilities.
Those details need to drive the purchasing decision. We should ask vendors to show us exactly what is logged, how long those records are available, and whether we can export them into our existing security tools. An audit log showing that someone signed in is useful, but it may not answer a question about a particular upload or action. We need to understand that difference before an investigation depends on it.
We should also know who can enable connectors, which internal resources the service can access, and whether employees can create agents that act on their behalf. Those capabilities need owners and boundaries. They should not become available to the entire organization simply because they were included in the license.
This is familiar security work. We already expect to manage accounts, review permissions, investigate activity, and remove access when someone leaves. AI should be brought into those processes.
There is a responsibility to employees here, too. Give them a useful tool and explain what information they can put into it, what uses require approval, and what activity the organization records. They should have a reasonable way to ask for additional capabilities rather than having to work around a service that does not meet their needs.
Then follow through. Deploying the tool without configuring it, assigning an administrator, or making the logs available to the security team leaves much of the work unfinished. Someone needs to own the service, review changes, and know how to retrieve evidence when a concern arises.
An enterprise platform will not automatically eliminate every unauthorized AI service. The organization still needs to identify other use, address it, and understand why employees continue to seek alternatives. But providing a managed option establishes somewhere for that work to go.
The staff are already making decisions about AI. It is time for the organization to make some of its own: provide an approved service, configure it deliberately, and establish the visibility needed to manage its use.
Don’t forget to have good policies, which will be in the next blogpost in this series.
Written by: Jonathan Kimmitt
Blog Jonathan Kimmitt
I think one of the more important conversations we need to have about AI Governance is also one of the simplest: Are we being honest about how and why we ...
Blog Andrew Hickman
Blog Jonathan Kimmitt
Copyright 2019 Cyber Security Design Concept by <a href="http://qantumthemes.com?rel=demo" target="_blank">QantumThemes</a>.