Listen to the article
Criminal services are pooling stolen accounts, API quotas and cloud resources to sell discounted access to AI models, with unauthorised use leaving some companies facing bills of up to US$1 million.
Criminal marketplaces are expanding beyond the sale of stolen passwords and personal data to offer access to AI accounts, API quotas and cloud computing. Buyers can also find proxy services that route requests through compromised credentials, giving them a way to use commercial models without paying providers’ standard prices.
Google’s Threat Intelligence Group describes a specialised supply chain behind the trade. Some operators break into accounts or systems to obtain credentials; others run the infrastructure, advertise access and manage customers. One group of illicit services, known as LLM Mixers, pools stolen credentials and computing resources, then routes subscribers’ requests to AI models.
The credentials may be taken from browsers, developer environments and cloud systems. Google’s investigations describe attackers using exposed API keys, stolen session tokens, cloud identity credentials and tokens associated with code-hosting platforms. Separately, TechRadar reported that a campaign targeting GitHub repositories exposed AI service keys alongside cloud credentials and other sensitive data, illustrating how development tools can become a source of stolen access.
The market’s offerings include verified accounts, token quotas and discounted plans, rather than just individual keys. Okta’s analysis of a 7GB infostealer dataset, published on Telegram in August 2026, found logs from 5,871 infected computers across 162 countries. The company identified thousands of valid session tokens for major technology and AI providers, as well as dozens of API keys stored in plain text. It also found that stolen authentication data can be used to bypass ordinary sign-in processes, including some multi-factor protections.
Demand has helped push prices upwards. Google reported that underground prices for some AI coding subscriptions and accounts more than doubled during 2026. VentureBeat likewise reported increased buyer interest and more sellers advertising AI accounts, while noting that the credentials can offer access not only to paid model usage but also, in some cases, to source code, data or computing infrastructure.
The financial consequences can be severe for the organisations whose accounts are compromised. Google and Mandiant said they had responded to incidents in which attackers used cloud environments and AI quotas without permission. In some cases, charges exceeded US$100,000 in a day; the most serious incidents reached US$1 million before they were brought under control. Palo Alto Networks’ Unit 42 has also warned that credentials can be fed into proxy services rapidly, allowing unauthorised use to generate substantial bills.
The trade is not limited to outright theft. Unit 42 describes intermediary services that sell AI capacity below ordinary retail prices, handling authentication, billing and the routing of requests across models. Some discounted offers may rely on promotional deals or arrangements that breach providers’ terms, but services that resell access cheaply need a low-cost supply of tokens to remain viable. Stolen credentials can provide one such source.
The expanding market also reflects a wider challenge for organisations: keeping track of credentials held by people, software and automated agents. Google’s reporting describes criminal operators exploiting compromised environments to monetise AI access, while separate incidents cited by VentureBeat show how exposed dashboards or poorly controlled machine credentials can create openings. For companies using AI tools, protecting API keys and session tokens is therefore not only a matter of preventing data theft; it can also help avert unauthorised usage and rapidly escalating cloud bills.
Source Reference Map
Inspired by headline at: [1]
Sources by paragraph:
Source: Fuse Wire Services


