AI Agents Are Hacking Retailers for $25 a Pop
Autonomous AI agents are automating attacks on online retailers, with an estimated cost of just tens of dollars per compromised company.
A new attack campaign is using autonomous AI agents to hack online retailers for just tens of dollars per company. This isn't a theoretical risk; it’s an active operation that shows offensive AI is now cheap, accessible, and highly automated. The key takeaway for builders is that the cost and skill required to launch scaled attacks have collapsed.
How does the attack work?
The campaign, detailed in a new report from Gambit Security, uses a swarm of coordinated AI agents to automate the entire attack chain. The operator didn't build a single super-agent. Instead, they used three distinct harnesses named Strix, Cairn, and Hermes, each with a specific job: reconnaissance, analysis, and exploitation.
The whole operation was orchestrated through API calls to various language models via OpenRouter. This detail is critical. The attacker didn't need to host or fine-tune models. They used a commercial routing service to access the best tool for each step, paying pennies for API calls that resulted in compromising at least 27 companies in under a week. The total cost was estimated at around $25 per successful breach.
What's the real implication for builders?
The use of OpenRouter is the most significant part of this story. It proves that the supply chain for offensive AI is now a commodity. Anyone with an API key and a few dozen dollars can orchestrate multi-agent systems to find and exploit vulnerabilities at scale. This isn't about a breakthrough in AI capabilities, but a breakthrough in accessibility.
This attack pattern shifts the economics of security. Defenses that rely on attackers having limited time, resources, or skill are now obsolete. Your web-facing services are not being probed by a human in a terminal, but by a relentless, automated system that is constantly learning and adapting its approach for the cost of a coffee.
The Verdict
This is a real inflection point. The barrier to entry for sophisticated, automated attacks has been obliterated. It's no longer about whether someone will find your vulnerability, but how quickly an automated agent will.
If your security posture assumes attackers operate on a human timescale, you need to reassess immediately. Defenses must move towards automated threat detection, aggressive rate-limiting on all endpoints, and assuming you are always the target of a cheap, scalable agent swarm. This is the new baseline.