For years, the small business owner in Penang or Cebu has relied on a quiet, unspoken form of security: obscurity. Who would bother hacking a family-run online furniture store when there are banks to rob?
The logic was not wrong. Breaking in took skill, time and patience, and skilled criminals price their time like consultants. They went where the money was.
That logic died this month, and it cost about US$50 to kill it.
Also Read: Singapore Polytechnic launches CASTLE to help SMEs strengthen cyber defences
According to a report by NorForge researcher Snorre Fagerland, first reported by Reuters on 19 October, a single Chinese-speaking operator used an open-source agent platform called Cairn to break into the websites of more than 100 companies, including Deere, SharkNinja and Cummins. The agents attempted more than 1,200 sites.
The operator’s instructions ran to fewer than 100 Chinese characters. The agents did the rest, taking a little under 23 hours per successful intrusion. Fagerland estimated the whole campaign burned roughly US$5,000 in tokens, or about US$50 per break-in.
“It’s not script kiddie stuff,” Fagerland told Reuters. Quite. It is something worse: script kiddie economics with professional-grade output.
Not the giants. The middle
The headlines named the American giants because giants make headlines. The more instructive detail sits lower down. NorForge found that the busiest cluster’s typical target was an online shop with traffic of between 200,000 and 1.5 million, in the US, Canada, Japan and Australia. The goal was to plant a web shell for lasting access to shop and payment systems, to be cashed in later.
Read that profile again with Southeast Asian eyes. It describes the region’s mid-sized direct-to-consumer brands: the beauty labels, gadget resellers and fashion startups that grew up on marketplace traffic and are now building their own storefronts.
Southeast Asia wasn’t on this operator’s list. That is luck, not design. Nothing about the method is tied to geography, and an operator already working through Japanese checkout pages will not find Bahasa Indonesia or Thai ones any harder.
The structural problem is that the region’s digital economy grew faster than its defences. An ERIA study cited by Positive Technologies found that only 68.5 per cent of small businesses in ASEAN use specialised software to protect their infrastructure. Roughly three in ten have nothing beyond whatever their hosting provider bundles for free.
The plugin you updated last Thursday
If that sounds abstract, consider the other security story of the week. Between 15 and 16 October, the distribution process for Rank Math PRO, a popular WordPress SEO plugin, was compromised. Some copies of version 3.0.123 carried code capable of creating persistent access to the sites that installed them. Administrators were told to treat their websites as potentially breached and restore from clean backups.
Also Read: Southeast Asian tech leaders are learning to trust AI agents, but not with production
This is how small businesses actually get hurt. Not through elaborate zero-days, but through the chain of plugins, themes, freelance developers and forgotten admin panels that holds a typical SME website together.
The founder does not know which version of which plugin is live. The agency that built the site moved on two years ago. The admin password lives in a WhatsApp group.
AI agents are tireless at exactly this kind of work. A human attacker gets bored scanning 1,200 storefronts for a misconfigured control panel. An agent does not get bored. It does not ask for a salary either.
Enterprises get guardians, SMEs get a blog post
The industry’s response so far tells you who matters. This week Microsoft unveiled Guardian Agents, which supervise enterprise AI agents in real time and treat them as potential insider threats, with task-scoped permissions and human approval for high-impact actions.
It is sensible engineering. It also sits inside Microsoft 365 E7, which is not a product line the kuih supplier in Shah Alam is likely to be shopping in.
Meanwhile, the bill for getting it wrong keeps climbing. IBM’s 2026 Cost of a Data Breach Report put Southeast Asia’s average breach cost at a record US$4.12 million, 12 per cent higher than last year, and IBM partly blamed AI for making attacks faster and cheaper to run.
That figure comes from just 26 organisations, so treat it as a direction rather than a price tag. But the direction is unmistakable. And a small business does not need a US$4 million bill to be finished. A week of hijacked checkout pages and a few hundred leaked card numbers will do the job.
So the region is drifting into a two-tier security economy. Large companies will buy AI to fight AI. Everyone else will rely on hope and a free SSL certificate. When attack costs fall to US$50 and defence stays priced for enterprises, the gap is not technical. It is a market failure.
What a sane response looks like
Market failures are what platforms and governments are for, and three moves would matter more than another cyber-hygiene awareness campaign.
Also Read: From KYC to KYA: how AI agents are reshaping payment risk
First, push security down the stack. The marketplaces, website builders, payment gateways and cloud resellers that host Southeast Asia’s small sellers are the only players able to defend millions of merchants at once. Automatic plugin patching, default multi-factor login and alerts for exposed admin panels should be the baseline, not a premium tier. Payment providers have the clearest incentive of all: the Cairn operator was after payment systems.
Second, treat agent infrastructure as a loaded weapon left on the table. NorForge saw this campaign only because operators had left their Cairn control panels open on the internet, and those panels leaked hundreds of credentials and API keys belonging to victims. Hosting providers in the region should scan for and take down malicious orchestration panels as routinely as they remove phishing pages.
Third, hold the AI labs to account for where their tools end up. NorForge’s data showed tasks running on a worker labelled after Anthropic’s Claude Code and DeepSeek’s low-cost V4.1 Flash model, though the firm offered that as a reading of the label, not a confirmed stack. Whatever the precise combination, model providers selling into ASEAN should publish regional abuse data and fund defensive tools for the small companies their products are now being turned against. Regulators signing AI safety agreements with these labs should be asking for exactly that.
Obscurity was never a strategy
Southeast Asian founders have long been told that cybersecurity is a cost centre to sort out after product-market fit. That advice assumed attackers were scarce and expensive. They are now neither.
The uncomfortable lesson of the Cairn campaign is not that one hacker got into Deere. It is that one hacker, armed with a few thousand dollars of tokens and a sentence of instructions, could plausibly work through every mid-sized online shop in Jakarta while asleep.
Also Read: The new ransomware playbook: Why ASEAN banks are losing the disclosure war
The region’s small businesses were never safe because they were hidden. They were safe because they were not worth the effort.
At US$50 a door, everyone is worth the effort.
The post AI agents broke into 100 firms for pocket change. SEA’s online shops fit the target profile appeared first on e27.