Cybercriminals are hijacking corporate cloud and artificial intelligence credentials, then reselling that access through underground subscription services that can generate fraudulent cloud charges exceeding $100,000 in a single day, according to a new warning from Google Threat Intelligence Group. The financially motivated campaigns, which GTIG calls "LLMjacking," have in some cases piled up cumulative charges reaching $1 million before victim organizations discovered the compromise and revoked access. Companies deploying large language models in production may be unknowingly paying for someone else's AI business.

The monetization scheme centers on what GTIG calls LLM Mixers, criminal routing gateways that bundle stolen API credentials and resell model access to underground subscribers. These services act as proxies, distributing requests across compromised accounts to obscure the original source of access. GTIG's analysis draws from multiple incidents in 2026, spanning a variety of cloud and LLM providers across North America, South America, and Asia Pacific. One threat actor, UNC6993, used tools including new-api and gcli2api to aggregate stolen keys and distribute requests across accounts. Another operation used stolen third-party API keys to power Jet Night, a legitimate token-routing service that executed more than 500,000 LLM API calls in under eight hours. That activity passed nearly $1 million in token costs to the victim organization. Jet Night advertises unlimited access to frontier models for approximately $60 a month, substantially lower than the typical cost of comparable legitimate access. GTIG suspects that occasional outages at Jet Night were caused by victims discovering compromises and revoking keys.

According to the report, initial access involves exposed or stolen cloud identity tokens, including IAM credentials, service account keys, GitHub personal access tokens, and active session tokens. Attackers also harvest LLM API keys from public repositories and scan for exposed model endpoints. Attackers were seen validating stolen keys with minimal single-token requests, sometimes using max-tokens=1. Such micro-transactions can confirm whether credentials work while remaining below conventional consumption-based detection thresholds. After establishing access, attackers enumerate cloud services, repositories, and data stores for additional credentials. They may escalate privileges, create unauthorized service accounts, inject SSH keys, or establish persistence through compromised workloads. GTIG observed a follow-on threat actor exploiting a user ID rebinding vulnerability in the LiteLLM gateway to elevate ordinary virtual keys into administrative credentials.

The report explains that selling stolen model access gives criminals a reason to keep that access working for as long as possible, which makes the quiet activity before the spending spike especially important. By the time someone asks why the cloud bill has exploded, the attacker may already have paying customers. "LLM Mixers turn stolen enterprise model access into a service for paying customers, while the victim's systems continue accepting requests through credentials they recognize," said Piyush Sharma, co-founder and CEO at Tuskira. Control of the routing layer could open access beyond the original compromised credential, the report warns. Testing a stolen key with a single-token request barely registers as consumption, but analysts shouldn't have to pull those connections together manually while the attacker gets a head start.

GTIG recommends tighter credential management, least-privilege permissions, granular API key scopes, monitoring for suspicious low-volume requests, and spending limits on model consumption. It also urges organizations to monitor unauthorized AI service enablement and gateway vulnerabilities. Organizations must restrict credentials to their intended models, workloads, and endpoints. The gateway compromise described in these incidents deserves particular attention, the report says, warning that a key issued for one application shouldn't become a reusable pass to other models and workloads. The subscription model for stolen AI access shifts the incentive structure for attackers, favoring sustained stealth over quick exploitation. Security teams that rely solely on consumption spikes may miss the validation and enumeration phases that set up the eventual bill shock.