blog

Security Operations

AI Agents

Change Management Used to Be a Weekly Meeting. Now It's Every Second.

Change Management Used to Be a Weekly Meeting. Now It's Every Second.

Yehonatan Rubin

|

|

Reading Time:

4 min

min

Table of Contents

Change management was designed around a simple assumption: changes are rare enough for people to review them.

A team proposed a change. It went to a change advisory board. Someone checked the risk, someone signed off, and the change went out in a window. Slow, yes. But it matched the pace of the work.

That assumption broke this year.

The CAB was built for human speed

A weekly review works when a team ships a handful of meaningful changes a week. Each one gets real attention. The reviewers know the systems. The risk conversation is short because everyone has context.

Now look at what lands in a security team’s queue today. Firewall rule requests. Access grants. New SaaS tools. Cloud permissions. Vendor integrations. Detection rule updates. Each one is a change. Each one carries risk.

AI multiplied the changes, not the reviewers

AI made every engineering team faster. Code assistants write more code. Infrastructure gets generated from a prompt. AI agents request their own access, connect to new systems and change configurations as part of doing their job.

The volume of change went up by an order of magnitude. The number of people who can approve a change did not move. It is still the same five people. More decisions. More approval fatigue.

When reviewers drown, one of two things happens. They rubber-stamp, and risk slips through. Or they block, and the business routes around security. Approvals move to Slack threads and DMs. Nobody records them. Then audit season arrives.

Every change is a security decision

Here is the reframe. A change request is not paperwork. It is a security decision with a deadline.

To decide well, the reviewer needs the same context every time. Who is asking. What asset is touched. What policy applies. What we decided the last time someone asked for this. What the live threat exposure looks like right now.

In most teams, the reviewer gathers that by hand. That is the actual cost of change management. Not the approval. The hunting before it.

Same rules, applied every time

TReAT handles change requests in the same queue as alerts and findings. Every request arrives already assembled: requester, asset, policy, precedent and threat context, in one case.

Most of your change rules already exist. They live in a wiki page, a policy document or a senior engineer’s head. TReAT drafts the rule your team already follows. You approve it. Then the agents run it the same way on every request, at 3pm or 3am.

Most change requests are routine knocks. A few are threats in a costume. Routine changes that match an approved policy move without waiting for a meeting. Changes that do not match go to a person, with the evidence and a recommendation already written. The reviewer stops hunting and starts deciding. Security moves from wait to done.

Then a new kind of change shows up. A new vendor review flow. A new class of AI agent asking for access. It does not need a new project. Same agents, same context. A new rule, approved by you.

Nothing changes until you say so

Speed in change management is only useful if it is trustworthy. So the guardrails are part of the design, not a footnote.

Agents run in shadow first, against your real queue, until you decide what graduates. You set what always needs a person. Every step is documented and reversible. The record shows what ran, in which tool, under whose permissions, and who approved it.

That record does two jobs. It proves the work to an auditor. And it teaches a new analyst how your best reviewer would have decided.

Change will not slow down again. The meeting cannot keep up. The decision can.