“The Most Significant Shift in Security… Since the Move to Cloud”

CEO and Founder, KineticSkunk

I don’t know about you but my news diet has been going keto on AI lately. And I don’t know whether the appropriate response is a new security policy or a bunker. But let’s leave the apocalypse planning to somebody else.

Right now, there is a much more immediate security consideration already sitting inside enterprise AWS environments because AI agents are basically able to make decisions across infrastructure without waiting for a human to approve their every move. AWS itself has called this “the most significant shift in security posture since the move to cloud”.

So today’s most pressing question for cloud teams is a bit less cinematic than “are we all gonna die when Skynet diverts our drinking water to cool its data centres”, but much more useful: if your AI agent can act inside AWS, who decides what it is allowed to do? And then what happens when the thing interacting with your AWS environment can make decisions at machine speed?

Your new colleague never sleeps

AWS has been moving rapidly in this direction itself. AWS DevOps Agent became generally available in March 2026, and was described as an “always-available operations teammate” capable of correlating telemetry with code and deployment information. The DevOps Agent could also identify operational improvements across AWS, in both multicloud and on-premises environments.

Since then, AWS has expanded the agent’s capabilities into release management, where it can do things like review code changes and assess dependencies, as well as look for potential drift from internally agreed standards before software reaches production.

Make no mistake, it’s incredibly useful technology, but it changes everything. Traditional controls were designed around relatively predictable software and how people used it. But agentic systems can choose tools dynamically. They can also chain actions together and behave differently depending on context.

AWS’s security guidance makes the implication fairly explicit: “Every agent needs its own identity with temporary, scoped credentials rather than persistent, broad access.” And because the agent operates quickly, a badly configured permission can become a much bigger problem much faster.

The old principle of ‘least privilege’ is suddenly more important than ever, and for reasons we couldn’t have predicted when it was first established.

Machine-speed operations need machine-speed visibility

There is a big, steaming pile of irony here. AI agents can help solve one of cloud operations’ oldest problems: it’s impossible for humans to monitor everything, all the time. But AI agents can begin investigating an incident before somebody has finished typing “Anyone else seeing this?” into Slack.

AWS says its DevOps Agent can reduce investigation time by bringing together information from observability tools, runbooks, repositories and deployment pipelines. Wonderful. But as automation becomes more capable, the quality of the actual operating model around it matters even more.

Who decides what an agent is allowed to do?

Which actions require human intervention?

What happens when an agent behaves unexpectedly?

And perhaps most importantly, can you see exactly what it did after the fact?

AWS’s September 2026 agentic security guidance says that one-off security assessments are no longer sufficient. Especially for these kinds of workloads. Ultimately, detection and response need to operate continuously because the systems themselves are dynamic.

That should sound familiar to anyone who has been following our argument about AWS operations.

At KineticSkunk, we’ve accepted the fact that the environment keeps changing, and will keep changing, accelerating even, at breakneck speed. But what underpins our philosophy, and what we insist on telling our clients, over and over again, is that operating discipline absolutely has to change with it.

So here’s a little word of advice: Treat an AI agent like a brilliant new employee who can work at 3,000 kilometres an hour, 24/7. Give it the access it needs, document what it owns. But maybe don’t hand it the master keys on day one.

Governance has a speed problem

AWS cites a massive gap in its current agentic-security guidance. AI agents are starting to interact with the same systems that cloud teams have spent years carefully securing. So they inherit many of our existing cloud-security problems, but they’re also introducing an important difference: speed.

Automation can take a huge amount of repetitive investigation and response work off engineering teams. That is the exciting part. But giving AI more autonomy without improving your operating model is basically hiring a Formula 1 driver and forgetting to build the brakes.

So, yes, there are some very big conversations happening about where AI might eventually take us, and they deserve serious attention. But while everyone else is debating bunkers and Skynet, there is a considerably more immediate question for anyone operating AWS today:

If your newest team member can think, act and move faster than any human in the room, are you equally clear about its parameters?

Because while the AI apocalypse may or may not arrive, the governance problem already has.

Sources: AWS Security Blog on agentic security and continuous detection and response, AWS DevOps Agent launch and documentation and AWS guidance on secure agent access patterns.