The Conversation That Felt Familiar
A few days ago, I was sitting in my office reading through the latest coverage of the Trump administration’s AI cybersecurity executive order. Like most people in cybersecurity, I wasn’t particularly surprised by the order itself. Governments were always going to get involved once artificial intelligence started crossing into national security, military planning, intelligence operations, and critical infrastructure.[1]
What surprised me was how quickly the conversation went somewhere else.
Within hours, security professionals, journalists, analysts, and technology executives were doing what they always do when a new piece of policy drops. They started debating the details. Was the voluntary framework strong enough? Was government oversight too aggressive? Should frontier models be regulated differently? Did the order go far enough, or did it go too far?[2]
I read article after article and found myself having the same reaction.
We’re talking about the wrong thing.
That doesn’t mean the executive order is unimportant. It matters. The fact that governments now view advanced AI systems as a national security issue is significant.[1] Ten years ago, that conversation would have sounded like science fiction. Today, it’s federal policy.
What struck me was how much of the discussion revolved around the models themselves. The entire debate seemed focused on what the models could do, who was building them, and how they should be evaluated before release.
The more I thought about it, the more familiar it all felt.
I’ve seen this movie before.
Not AI specifically. The pattern.
Every Technology Revolution Looks Different Until It Doesn’t
I’ve spent more than three decades in technology and cybersecurity. I’ve been around long enough to remember when Linux was the disruptive technology that everyone argued about.
Back then, every conference hallway had somebody explaining why their distribution would dominate the future. Every magazine article seemed filled with predictions. Every expert knew exactly how the story would end.
Most of them were wrong.
What I remember most wasn’t the technology itself. It was the certainty.
People were absolutely convinced they knew where things were heading.
Today, I hear that same certainty whenever AI comes up.
Ask ten experts which model will dominate government systems five years from now and you’ll get ten different answers delivered with complete confidence. Some will tell you Claude is the future because of its coding capabilities. Others will argue that ChatGPT’s versatility makes it the safer long-term choice. Open-source advocates are convinced proprietary systems eventually lose. Proprietary vendors are convinced open-source models can never catch up.
The truth is nobody knows.
AI is still young.
The technology is advancing at a breathtaking pace, but we’re still in the early chapters of the story. History is littered with examples of dominant technologies that looked unstoppable until they weren’t.
That’s one of the reasons I find the current debate so interesting. We are investing enormous amounts of energy trying to determine which model wins, while something much bigger is happening underneath us.
The models are becoming infrastructure.
The Moment I Stopped Worrying About Models
Somewhere over the last year, I found myself paying less attention to the model wars and more attention to what people were connecting those models to.
At first, it was simple integrations.
Then came agents.
Then orchestration frameworks.
Then workflow automation.
Then MCP servers.
Then autonomous systems capable of moving information between multiple environments without direct human involvement.
And that’s when something clicked for me.
The model was becoming the least interesting part of the equation.
I recently watched a demonstration of an AI system connected to multiple enterprise platforms. The demo was impressive. The agent could retrieve information, query databases, update records, open tickets, interact with APIs, and coordinate actions across multiple systems.
Everyone watching seemed fascinated by how intelligently it performed the tasks.
I kept thinking about something else entirely.
What permissions does it have?
That’s probably thirty years of cybersecurity damage talking.
Attackers don’t care how smart something is.
They care what it can access.
The more I watched organizations build AI-powered workflows, the more familiar everything started to feel. An AI model sitting in a browser window doesn’t concern me very much. An AI agent connected to cloud infrastructure concerns me. An AI agent connected to source code repositories concerns me. An AI agent connected to government systems, military platforms, intelligence environments, financial networks, healthcare infrastructure, or operational technology concerns me a great deal.
Not because of the model.
Because of the authority.
People keep asking whether artificial intelligence is dangerous.
I think that’s the wrong question.
A chainsaw sitting on a shelf isn’t dangerous.
A rifle sitting on a table isn’t dangerous.
Danger appears when capability is combined with authority.
The same principle applies here.
The model isn’t the story anymore.
The authority we’ve attached to it is.
The New Attack Surface Nobody Wants to Talk About
One of the things incident response teaches you is that attackers are remarkably consistent.
Security professionals love talking about new threats, but most breaches follow familiar patterns. Attackers look for trust relationships. They look for privilege. They look for access. Then they figure out how to turn those things into leverage.
The technology changes every few years.
The attack paths rarely do.
That’s why I find the rise of agentic AI so fascinating. For all the discussion about model capabilities, what we’re really watching is the creation of a massive new layer of identities.
Every AI agent becomes an identity.
Every automation becomes an access path.
Every MCP integration creates a trust relationship.
Every trust relationship creates another opportunity for things to go wrong.
NIST has spent years emphasizing concepts that eventually became associated with Zero Trust, specifically the idea that trust should never be assumed and access should be continuously validated.[3] Strip away all the marketing and that’s still one of the most important ideas in cybersecurity.
Trust is expensive.
Authority is dangerous.
Privilege always creates risk.
Those principles don’t disappear because AI arrives.
If anything, they become more important.
OWASP’s research into LLM security highlights risks including excessive agency, excessive permissions, insecure tool usage, and prompt injection.[4] When I read those categories, I don’t see AI problems.
I see identity problems.
I see authorization problems.
I see privilege management problems.
I see the same issues we’ve been fighting for decades wearing a new coat of paint.
The Real Challenge Is Still Human
What genuinely worries me isn’t some Hollywood scenario involving self-aware machines.
I’ve spent too much time dealing with human beings to believe artificial consciousness is the biggest problem on the horizon.
What worries me is a government agency, military organization, or enterprise connecting increasingly capable AI agents to increasingly sensitive systems while convincing themselves they’ve solved the problem because the model passed a security review.
A model review tells you something about the engine.
It tells you very little about the vehicle somebody is building around it.
The executive order focuses on understanding frontier models and their national security implications.[1][5] That’s important work. Governments absolutely need to understand what these systems are capable of.
But eventually, the conversation moves beyond capability.
Eventually, the conversation becomes about authority.
Who gets access?
Who grants access?
What systems can the agent touch?
What actions can it take?
What happens when something goes wrong?
Those questions are harder.
They’re also far more important.
The recent investment flowing into AI observability, runtime monitoring, agent security, and governance platforms suggests that investors are beginning to recognize the same reality.[6][7][8][9] The market is slowly realizing that deploying AI is relatively easy.
Controlling AI is much harder.
Where This Ends
If thirty years in cybersecurity has taught me anything, it’s that technology rarely creates entirely new problems.
More often, it magnifies existing ones.
AI is not changing the fundamental laws of security.
Attackers still want access.
Privilege still matters.
Trust relationships still create risk.
Identity is still the battleground.
The executive order is an important step. It acknowledges that advanced AI systems are now part of the national security conversation.[1]
But the real challenge isn’t understanding what AI can do.
The real challenge is controlling what we’re willing to let it do.
The future isn’t just AI.
The future is AI with access.
The future is AI with authority.
The future is AI connected to systems capable of influencing real-world outcomes at machine speed.
And if history is any guide, the organizations that thrive won’t necessarily be the ones with the smartest models.
They’ll be the ones that understand a simple truth we’ve been learning over and over again for decades.
Technology creates capability.
Authority creates consequences.
About the Author
Len Noe is a Solutions Architect at BeyondTrust, Transhuman, Podcaster, International Cyber Security Speaker, Author, Technical Evangelist, and Biohacker with 11 implanted microchips. A former blackhat and ex-1% motorcycle club member with more than 30 years in technology, he has presented in over 70 countries and is featured in the documentary I Am Machine, which premiered at DEF CON 2025.