Between late September and early October, an attacker broke into seven South Korean financial institutions and exfiltrated the personal data of about 68,000 people. CrowdStrike traced the campaign to ARTEX, an open-source AI hacking agent built in China and run alongside Anthropic’s Claude Code, and said the actor could be a 26-year-old. President Lee Jae Myung warned that AI can now make hacking “easy for even those without special skills.” An official at the Korea Financial Security Institute added, “It is correct that AI was used in the attack, but the AI did not act autonomously without human involvement — a hacker used the AI as a tool.”
Look closely at that hacker. CrowdStrike found no sign of an elite team or a state program, only an actor who paired agentic tools with established techniques. The human contribution looks thin: choosing Korean lenders as targets, aiming the agents at them, and asking Claude where the stolen data might be sold. The software supplied most of the technical legwork. The operator supplied the intent.
This is not the first example of state-backed groups leaning heavily on AI. In November 2025, Anthropic disclosed that a group it assessed with high confidence to be Chinese state-sponsored used Claude Code to run an espionage campaign against roughly 30 organizations, with the AI doing an estimated 80–90 percent of the tactical work. People still selected the targets and signed off before reconnaissance moved to exploitation, and the AI “frequently overstated findings and occasionally fabricated data.”
When attackers have AI at their disposal, their expertise multiplies. One individual with agentic tools can now reach multiple institutions in a matter of weeks, which historically required a skilled team. Meanwhile, defenders meet that threat with a talent gap that has persisted for years. The people who can judge which of a thousand machine-generated findings is a real attack path remain scarce.
I have seen this from the inside
Before I co-founded Synack in 2013, I was an operator in computer network operations at the National Security Agency. We had extraordinary tools. None of them decided what was worth going after, recognized which opening justified the risk, or improvised when a target behaved in a way no playbook anticipated. That was the operator’s job, and it was the job that decided whether a mission succeeded. The best operators I worked alongside were rarely the ones with the best tools. They were the ones who saw a path nobody else had thought to look for.
That is creativity, and it is the thing AI supplies least. A model generates from patterns it has already seen. It can produce a thousand variations at machine speed, and some of them will be good. What it cannot do is want an outcome, invent an objective nobody has written down, or sense that one odd behavior in a forgotten partner portal is the entire game. In Korea and in the state-backed campaign, a person decided where the machine should point.
People make the machine better
The evidence runs in one direction. In research published in June, academics benchmarked six frontier AI systems against realistic corporate networks. The best performer, working on its own, succeeded on 16.1 percent of web attack tasks. Given more concrete guidance, the kind an experienced operator supplies, its success rate roughly doubled, to 33 percent. The machine was not the bottleneck. Direction was.
That is the bet behind the merger of Synack and NetSPI, which completed in October. Both companies spent years building deep benches of offensive security talent, including the Synack Red Team, and the market keeps asking whether AI makes that talent obsolete. The Korean case points the other way. AI is lowering the cost of attacking, which raises the value of defenders who can judge what matters. It lets one of those experts aim their judgment across far more work than any person could reach alone.
Headcount is the wrong scoreboard
This matters well beyond security. Too many executives still judge AI by how many people it lets them cut. Labor savings are easy to count, which is why they dominate business cases, and they miss the larger prize.
If AI strips away the routine hours so your best specialist can take on twenty hard problems instead of ten, you have not eliminated a role. You have doubled the reach of the person whose judgment is hardest to replace. The companies that capture that value will redesign work around their most creative people, rather than bolting AI onto old processes and counting the empty chairs.
AI will commoditize plenty of tasks, but it will not commoditize the people who know which tasks matter. Korea’s attacker needed little technical skill and still had to decide where to aim. Defenders face the same decision across far more systems. The executives funding the defense should judge AI by how far it extends the people who make that call.
The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.