---
title: "Crossed Wires: AI flips the script on the economics of hacking and cybersecurity"
description: "A single operator can now direct AI hacking swarms to scan, exploit and ransom neglected devices worldwide, collapsing human processing limits and exposing decades of unpatched vulnerabilities."
type: "NewsArticle"
publisher: "Daily Maverick"
site: "https://www.dailymaverick.co.za"
section: "BUSINESS REFLECTION"
author: "Steven Boykey Sidley"
author_url: "https://www.dailymaverick.co.za/author/stevenboykeysidley/"
canonical_url: "https://www.dailymaverick.co.za/article/2026-09-20-crossed-wires-ai-flips-the-script-on-the-economics-of-hacking-and-cybersecurity/"
published: "2026-09-20T23:02:46"
lang: "en-ZA"
word_count: 1483
---

# Crossed Wires: AI flips the script on the economics of hacking and cybersecurity

> A single operator can now direct AI hacking swarms to scan, exploit and ransom neglected devices worldwide, collapsing human processing limits and exposing decades of unpatched vulnerabilities.

By Steven Boykey Sidley · Published 21 September 2026, 01:02 SAST

## Key points
- Decades of forgotten kit – from hospital controllers to old middleware and COBOL relics – survived because there simply weren’t enough skilled hackers to try every door.
- That scarcity is gone: cheap AI agents can swarm, probe and exploit at machine speed – examples include Taiwan’s 21-system mapping, mass credential theft and Anthropic’s “vibe hacking” campaigns.
- The economics have flipped: lone operators can rent or run autonomous attackers, black hats need only one hole while defenders must lock them all, and patch windows are being weaponised.
- Defence now hinges on automation – zero trust, moving-target architectures and treating “every byte as a burglar” – but the race to retool is urgent and far from guaranteed.

## Content

Some decades ago, I was on a business trip overseas. I was the chief technology officer for a large tech company at the time, and we were visiting a well-known cybersecurity product developer that made encryption hardware for set-top boxes (remember those?).

As part of the tour that had been arranged for us, we were given a presentation by one of the executives, an iconic crypto scientist responsible for one of the most enduring security algorithms on the internet.

In his talk he described and demo’d some of the more edgy cybersecurity wizardry that was being researched at the company. I remember thinking wistfully – what a cool thing to do, thinking up ways to keep digital secrets, ways to hack those secrets and then ways to armour them further. A skill available only to the smartest of scientists who seemed to live in a parallel universe of maths, code and algorithms, often lurking in dangerous digital neighbourhoods populated by skilled bad guys trying to do bad things, and trying to stay a step ahead.

Sounded like something I would have loved to have done, had I had the talent.

### How the game has changed

Consider a hospital basement in which there is a building-management controller (you know – air conditioners, lighting, energy, alarm systems, access control) running firmware last updated before the iPhone was a thing. It has an ugly web interface and a default administrator password because nobody has bothered to change it since installation. For 15 years it has been sitting on the open internet, unmolested. It is not secure. It is merely unvisited.

Unvisited, obscure, unnoticed. The world’s digital ecosystems have long enjoyed this under-appreciated form of protection – there simply haven’t been enough good hackers to attack all of it.

The digital world is a largely undisturbed archaeological site spanning nearly 70 years. Of course we have our current cloud-native applications, zero-trust networks, passkeys and security teams running state-of-the-art detection systems.

But underneath them lie decades of accumulated detritus – ossified applications written in older languages such as C (and mediaeval languages such as Cobol), neglected and creaky middleware, forgotten APIs, routers, firewalls, switches, network-connected printers, industrial controllers, WordPress plug-ins, authentication servers, third-party libraries and innumerable bespoke programs written by someone who retired in 2004 and whose memory is fading.

The reason this ecosystem has survived and thrived is because finding an obscure vulnerability, understanding the architecture around it and turning the weakness into a useful and profitable attack requires rare human expertise. Unauthorised penetration of these systems has always been artisan work – labour-intensive, highly skilled, expensive and fundamentally constrained by human speed.

It has been a well-understood landscape. The attack surface is vast. The black hats would primarily attack high-value targets where reward was greatest, and the white hats would build expensive systems to protect the most important perimeters – the banks, the military installations, the hospitals, the cloud providers. The war has raged back and forth for decades.

And now the coming swarm of artificial intelligence (AI) agents, which is quickly rewriting the terms of engagement.

### Digital teardrop

The twin rate-limiters of a massive attack surface and rare human expertise are collapsing in real time. Anyone with a medium grasp of how to build an AI agent system (not that difficult – I learnt how to do it in weeks) can now let loose a hacking swarm across any part of this attack surface with instructions to keep trying, prodding, probing endlessly (and autonomously) until a way to access everything and anything is found.

One does not need to know that much about the locks that are being picked – that’s the agent’s job. And one does not need to know much about hacking. The instruction would simply be – gain access any way you can, and tell me when you are in. The agent takes care of the deep technical details.

That is one of the spoils of agentic AI.

And so we are now seeing [reports](https://www.cnn.com/2026/08/13/tech/china-taiwan-ai-agent-cyberattack-intl-hnk) like this: over four days in July, AI agents mapped 21 Taiwanese government systems, cracked 85 accounts and extracted 2,500 personnel records. Taiwan’s digital affairs ministry noted that the agents exploited vulnerabilities in secondary systems, working faster, more cheaply and at far greater scale than had been seen before. Or the [just-announced](https://www.theguardian.com/technology/2026/sep/18/openai-hacked-anthropic-claude-chatbot?) Hacktron AI campaign, which used Anthropic’s Claude to penetrate systems associated with none other than OpenAI (note that this was a white-hat hack – the hackers informed OpenAI of the exploit and received a bug bounty).

In August 2025, a single criminal with limited technical skills ran a data-extortion campaign against 17 organisations in the space of a month, directing an AI coding agent to scan VPN endpoints, penetrate networks, harvest credentials and then, as a final flourish, compose bespoke ransom notes demanding up to $500,000. Anthropic’s threat intelligence team called it “[vibe hacking](https://www.anthropic.com/news/detecting-countering-misuse-aug-2025)”.

A few months [later](https://www.anthropic.com/news/disrupting-AI-espionage), a Chinese state-sponsored group, tracked as GTG-1002, used Claude Code against some 30 commercial and defence targets. And by early 2026 Amazon’s threat intelligence unit had found a single actor attacking internet-facing edge infrastructure across 55 countries, having wired Claude and DeepSeek models into an automated framework that scanned management interfaces, read the responses and chose its own exploit paths without anyone watching.

The important cybersecurity story is not simply that AI makes hackers’ lives easier. It is that AI is destroying the economics that have protected most software for decades.

An autonomous agent does not get tired – thousands of agents can be spawned that can probe thousands of systems simultaneously, continuously changing tactics and talking among each other as they discover what works.

Anthropic’s September 2026 [threat-intelligence report](https://www.anthropic.com/threat-intelligence-report-september-2026) describes sophisticated attacks no longer necessarily requiring much sophistication at all. It reports automated “exploit foundries”; agent swarms performing “reconnaissance”; and exploitation and data theft against multiple victims in parallel.

Tasks that previously required teams of skilled operators can increasingly be undertaken by lone individuals directing AI systems from a basement in a residential home.

Which has led to this startling [statistic](https://www.anthropic.com/research/glasswing-initial-update) – what took a hacker two years to do in 2018 now takes a few hours. That is WTF territory.

### White hats vs black hats: Who wins?

There is a standard industry line here, which is white-hat AI will protect us from black-hat AI. But it does not take much to question that theory. The first (and most obvious) is that black hats need only one door to penetrate to potentially cause a whole heap of trouble. White hats must protect all the doors – a much more complex undertaking.

And then there is this – it is in fact a crime for a white-hatter to go out and do this without permission of the owners of the systems they are trying to protect. They need approval, permission, compliance audits, contracts and the rest of the slow-moving human process sludge. Black hats ask no permission from anyone. They can move faster than white hats and much faster than enterprises.

The second flaw is that even when a vulnerability is uncovered, it is not always patched immediately, it is not always patched by everyone using the flawed system (and not always patched everywhere simultaneously). It is a simple matter for black hats to look out for security patch announcements and then to dash off to hack those who have not yet upgraded (how quickly do you upgrade your laptop or mobile operating system when notified?).

This leads to the obvious question – is computer security dead in the age of AI? There are in fact ingenious strategies under way to defend against what is already here. All of them involve removing the human from the loop.

These range from smart “zero trust architectures” where every internal subsystem and every movement of bits is considered a threat until a digitally signed “passport” is presented, to “moving target defence” architecture which constantly shifts the underlying execution environment (like changing memory configurations and API parameters) so that the hackers never encounter a static target (sort of like changing your door look every few minutes).

For decades we have scattered billions of digital doors around the world. Some are heavily fortified, some have sophisticated locks, some haven’t been opened for decades and some have been left unlocked. We were reasonably safe because there weren’t enough burglars to try every door. Now there are.

The solution seems to lie in treating every byte as a burglar. That’s how we win, if we can move fast enough. **DM**

*Steven Boykey Sidley is a professor of practice (ex-JBS, University of Johannesburg), a partner at Bridge Capital and a columnist-at-large at Daily Maverick, where he writes the weekly Crossed Wires column. His new book, It’s Mine: How the Crypto Industry is Redefining Ownership, is published by*[***Maverick 451***](https://shop.dailymaverick.co.za/product/its-mine-how-the-crypto-industry-is-redefining-ownership/?_gl=1*1rxvep4*_gcl_au*OTMxMTUwODIuMTc2ODMyODg4OQ..*_ga*MTc0MDU3MjcyNi4xNzUyNzE1NTQz*_ga_Y7XD5FHQVG*czE3NzE3NzMzNDYkbzc3NyRnMSR0MTc3MTc3MzM2NyRqMzkkbDEkaDM2NDEzNDA5NiRka09DakM4NE9CRHlrTTlvQ3JzWFNvLXg4bi0zMDJja2lPdw..)*in South Africa and the Legend Times Group in the UK/EU, available now.*