When the generative AI wave hit, I was working inside the cybersecurity research world, watching security roadmaps get rewritten in real time. I remember a conversation with a researcher at one of the frontier labs who told me that one of his biggest fears was that attackers would be able to use this technology to completely automate the kill chain. To him, the question was never if, but when. That got me thinking about what would happen to the economics of attack once they did.

Then, November 2025. Anthropic disclosed what it called the first AI-orchestrated cyber espionage campaign. A Chinese state-sponsored group (tracked as GTG-1002) used an agentic AI coding tool to run the intrusion chain itself: reconnaissance, exploitation, credential harvesting, lateral movement, exfiltration. It is estimated that the AI executed roughly 80-90 percent of the tactical work, targeting around 30 organizations, including tech companies and government agencies. Anthropic estimated human involvement at key phases was as little as 20 minutes of work. The humans picked targets and said “yes, continue.” The machine did the intrusion.

Some respected researchers question how much the AI really did on its own, citing limited technical detail in the report. But even the skeptical read concedes the point that matters. The workflow now exists. Capability curves move in one direction.

What happens to security when the attacker’s marginal cost approaches zero?

Move One: Offense Got Cheap

Full disclosure before the numbers. I sell in this market every day. And, yes, I get it: an AE at an AI security company arguing that AI security matters is like the barber telling you that you need a haircut. Discount my conclusions accordingly and check every number yourself. They hold.

According to the 2026 Hoxhunt Phishing Trends Report, AI generated phishing went from under 5% of detected attacks to 56% in a single month (December 2025), a 14x surge that has held into 2026. A controlled study found AI-generated phishing achieved a 54% click rate versus 12% for traditionally written emails. AI phishing tools can be purchased for as little as $75.

AI is collapsing the cost of attack in real time, and it isn’t doing so incrementally, but rather structurally. The skill floor continues to drop, and the volume ceiling has completely disappeared. Before AI, cyber criminals usually had deep technical expertise. Most ambitious plans often required significant resources to pull off. Bad news for defenders: AI has removed the technical and resource barriers to a successful attack, which can now be conducted at negligible cost and with minimal know-how.

Worth being precise about what collapsed. GTG-1002 was run by a state-sponsored group that built a custom orchestration framework. That is not zero-cost attacking, and the full cost of breaching a hardened enterprise, initial access, infrastructure, evasion, monetization, has not gone to zero. What collapsed is the marginal cost of a convincing attack and the skill floor for executing known techniques. The premium that expertise used to command is evaporating. That is the repricing.

Move Two: Defense Got Cheap Too

The cost of offense is falling. So is the cost of defense. I’m constantly reading doom articles about how attackers have the advantage due to AI, and yet they rarely mention that the same economics benefit defenders as well.

According to the 2025 IBM Cost of a Data Breach Report, organizations using AI and automation extensively in security saved an average of $1.9 million per breach and shortened response times by 80 days. Correlation is not causation, and organizations using AI extensively also tend to be better funded, more mature, and better staffed. But the direction holds across all five years of the report. Separately, global average breach costs fell in 2025, from $4.88M to $4.44M, the first decline in five years, credited to faster AI-assisted detection and containment.

The expensive human middle of security programs, triage, monitoring, reporting, compliance evidence-gathering, is exactly what AI automates best.

A skeptic may read those numbers and conclude that the defenders are winning. Look closer. IBM’s report covers breaches from March 2024 to February 2025. The AI phishing surge hit in December 2025, ten months after that window closed. Those falling costs describe the world before the wave arrived. And the decline is an average, which hides the sorting underneath it. US breach costs rose to $10.22 million. The $1.9 million in savings went to the minority of organizations running AI-driven security with real governance behind it. The average fell. The gap widened. That gap is the story.

When both sides of a conflict get cheap simultaneously, where does the value go?

Move Three: The Scarce Asset Moved

When execution gets cheap, the scarce asset becomes knowing what to execute, what to trust, and how to translate machine-speed risk into human-speed institutional decisions.

Markets abhor a vacuum, and when value drains from execution it has to pool somewhere else. Three things appreciate. First, the AI layer itself becomes the dependency. Second, governance becomes the differentiator, not the tooling. And third, judgement gets repriced upward.

Cheap defense at scale requires AI infrastructure, so now every cheap defensive deployment effectively runs on AI. The security of the AI becomes the security of everything downstream. AI security stops being a product category. It becomes the infrastructure layer everything else runs on.

Now that everyone can buy the same cheap AI defense, what’s lacking at the enterprise level is effective governance. 97% of organizations that suffered an AI-related breach lacked proper AI access controls. 63% of breached organizations had no AI governance policy at all. Shadow AI, employees using unsanctioned tools, factored into 20% of breaches and added $670K to average costs.

The gap between adoption and oversight is where breaches now live. The tooling still matters. But the tooling that matters now is the kind that makes governance possible: visibility into what AI is running, who’s using it, what it touches. You can’t govern what you can’t see.

The people who can do that translation, between the technical layer and the board, between the threat and the budget, become more valuable, not less.

What This Means for Three Readers

I spend most of my days speaking with CISOs, executives, and boards. A CISO told me last month that his budget conversations have completely changed shape. He used to talk about visibility into enterprise tooling, but now he spends his time asking about whether he can see and govern the AI that his organization already runs on, sanctioned or not.

In my discussions with executives, I’ve also noticed a rather similar shift in tone. AI security is no longer considered a simple line item under IT. It’s starting to be viewed as the dependency underneath every AI initiative that the board approves. Treating AI security as a line item is how you end up in the 63%.

The repricing also applies to the person who is building a career. While tasks continue to get cheaper, judgement, domain depth, and the ability to translate between technical reality and institutional decisions are getting expensive. We’ve all heard the cliché, “AI isn’t going to replace your job, but people using AI will.” I’d rephrase that to AI won’t necessarily replace your job, but those who use AI as a tool to sharpen their thinking, rather than outsourcing it, will replace those who don’t.

The Through Line

In 2015 I published in the Journal of Political Risk on the campaign against ISIL, an adversary that was part terrorist network, part guerrilla army, part proto-state. The question underneath the strategy debate was the one that has followed me since. What do institutions do when the threat stops fitting their doctrine?

The question took me to a global research and advisory firm, where I spent over a decade walking security leaders through cloud, ransomware, and the arrival of generative AI. I’m currently working for a Canadian cybersecurity firm, selling AI security. Same question, fifteen years, new economics.

What This Newsletter Does

Each edition of this newsletter will track the repricing as it happens. My job forces me to live at the front of this market. I sell in this market every day. I want this newsletter to read like it. I will focus on one development, one argument about where the value moved, and what it means for the people who must decide things. Security, AI, and the political economy underneath both. The first one’s done. You’re reading it.