AI-Driven Cyberattacks Target Online Retailers at a Cost of Just $25 Per Breach

The landscape of cybersecurity is undergoing a fundamental shift as the barrier to entry for sophisticated digital assaults continues to plummet. Recent findings from the Israeli security firm Gambit have illuminated a troubling trend: the democratization of high-level hacking through the use of autonomous artificial intelligence. In a striking demonstration of this new reality, a single threat actor managed to target 105 online retailers over the course of just five days, successfully compromising 27 of them. The most alarming aspect of this campaign was not just its success rate, but its staggering cost-effectiveness. On average, the attacker spent only $25 per company to orchestrate these breaches.

This development serves as both a dire warning for the global retail sector and a signal of a burgeoning, unethical "side-hustle" for individuals with moderate technical skills. By leveraging open-source AI frameworks, attackers are now able to automate the most labor-intensive aspects of a cyberattack—vulnerability research, exploitation, and campaign management—at a fraction of the cost and time required for traditional, human-led operations.

The Economics of Autonomous Exploitation

The financial data uncovered by Gambit’s research provides a granular look at the efficiency of modern AI-driven crime. By analyzing the attacker’s account balance on OpenRouter—a platform used to access various AI models—researchers found that the total expenditure over a four-week period ending August 25 was a mere $7,005. When distributed across the numerous targets pursued during that timeframe, the cost of an individual attack becomes remarkably low.

While the average cost sat at $25 per company, the price of entry varied depending on the complexity of the target’s defenses. For the most vulnerable retailers, the cost of a successful breach was as low as $3.13. Even for the most robustly defended targets in the group, the attacker only needed to spend $79.31 in AI processing fees to find a way in. This represents a radical shift in the return on investment for cybercriminals. When a breach that costs less than a tank of gas can yield hundreds of thousands of pieces of sensitive consumer data, the economic incentives for such attacks become nearly irresistible.

The efficiency of these incursions is further highlighted by the speed of execution. Gambit reported that most successful penetrations took only a few hours. In a traditional setting, a human hacker might spend days or weeks manually probing a corporate network for weaknesses, testing various exploits, and attempting to pivot through internal systems. AI removes these temporal bottlenecks, allowing for a high-volume, "spray-and-pray" approach that retains the precision of a targeted attack.

A Specialized Toolkit: Strix, Cairn, and Hermes

The technical foundation of this campaign relied on three specific open-source AI harnesses, each designed to handle a different stage of the cyberattack lifecycle. These tools—Strix, Cairn, and Hermes—function as a coordinated suite that effectively removes the human from the "attack loop."

Strix is utilized for the initial phase of the operation: vulnerability searching. Rather than a simple automated scanner that looks for known software versions, an AI-powered tool like Strix can interpret code and identify logic flaws or obscure vulnerabilities that traditional signature-based scanners might miss. It acts as the "scout," identifying the weakest points in a retailer’s digital perimeter.

Once a vulnerability is identified, Cairn takes over. Cairn is described as an autonomous end-to-end exploitation engine. It does not merely flag a hole in the security; it actively works to bypass defenses, gain entry, and establish a foothold within the target’s infrastructure. Because it is powered by large language models (LLMs) capable of understanding and generating code in real-time, it can adapt its tactics on the fly, reacting to security measures as it encounters them.

The third component, Hermes, serves as the orchestrator. It manages the broader campaign, coordinating the actions of Strix and Cairn across multiple targets simultaneously. Hermes allows a single operator to manage dozens or hundreds of concurrent attacks, providing a level of scale that was previously the sole province of well-funded state-sponsored groups or large-scale organized crime syndicates. By using OpenRouter to access the necessary computational power, the attacker was able to run these sophisticated models without needing to maintain an expensive private server infrastructure.

The Human and Financial Cost of Compromise

The results of this five-day blitz were devastatingly productive for the attacker. Gambit’s investigation revealed that 600,000 active credit card details were exfiltrated from just two of the compromised businesses. This volume of data, if sold on the dark web or used for fraudulent transactions, represents a profit margin that dwarfs the $7,005 initial investment.

In addition to direct data theft, the attacker successfully installed card skimmer scripts at five other retail sites. These scripts are particularly insidious because they sit silently on the checkout pages of legitimate websites, capturing customer payment information in real-time as it is entered. Unlike a one-time data breach, a skimmer can provide a steady stream of fresh data for as long as it remains undetected.

The researchers also noted that the attacker gained varying levels of access to an unspecified number of major companies. The implications of this are broad; even if credit card data was not immediately stolen, the presence of an autonomous agent within a corporate network can lead to the theft of intellectual property, the deployment of ransomware, or the compromise of employee credentials.

Gambit has taken steps to mitigate the damage by contacting all the affected companies. However, the sheer volume of targets and the speed at which they were compromised suggests that many retailers are ill-prepared for the velocity of AI-driven threats. The notification process itself is a challenge, as many organizations may not even be aware that a breach has occurred until the stolen data begins appearing in fraudulent transactions.

The Transformation of the Cybercrime Landscape

The intensity and success of these attacks signal a transformative moment in cybersecurity. For years, experts have warned that AI would eventually be used to automate hacking, but the Gambit research suggests that this future has already arrived. The transition from manual exploitation to autonomous AI agents represents a "force multiplier" for cybercriminals.

One of the most significant shifts is the level of sophistication now available to low-level actors. In the past, a "script kiddie" was limited to using pre-written tools against known vulnerabilities. To attack a major retailer required significant expertise. Now, with AI harnesses, an individual with basic technical knowledge can command a system that possesses the collective "intelligence" of the models it is trained on. This effectively allows an amateur to perform at the level of a professional penetration tester or a seasoned hacker.

This "human-out-of-the-loop" model of incursions is expected to become more prevalent. As AI models become more capable and the cost of API access continues to fall, the frequency of these attacks will likely increase. Organizations that rely on traditional security timelines—where a patch is applied weeks after a vulnerability is discovered—will find themselves increasingly vulnerable to agents that can find and exploit those flaws in a matter of hours.

The retail industry is particularly at risk due to the high value of the data it handles and the complexity of modern e-commerce platforms. Many online stores rely on a web of third-party plugins and integrations, each of which presents a potential entry point for an AI agent like Strix to discover.

A New Reality for Digital Defense

The Gambit report serves as a stark reminder that the tools of the trade for cybercriminals are evolving faster than the defensive strategies of many businesses. The fact that a campaign of this scale could be conducted for such a minimal cost suggests that the traditional "cost of doing business" for hackers has been permanently lowered.

As AI-driven attacks become the new standard, the security community is being forced to rethink its approach. If the attackers are using AI to find vulnerabilities, defenders must increasingly rely on AI to identify and block those attempts in real-time. The window for human intervention is closing, replaced by a high-speed environment where machine-on-machine conflict defines the frontline of digital security.

For now, the 27 compromised retailers stand as a testament to the efficacy of these new methods. As more IT workers and opportunistic actors discover the potential for these "unethical side-hustles," the pressure on the retail sector—and the broader digital economy—will only continue to mount. The era of the $25 hack is not just a possibility; it is a documented reality.