Recent investigations into LLMJacking have revealed a troubling trend where threat actors exploit vulnerabilities in cloud environments to monetize compromised resources. According to a report by Google Cloud, the primary method of initial access for these attackers has been through exposed or stolen cloud identity tokens, including IAM credentials and API keys. This exploitation allows them to infiltrate organizations’ infrastructures, leading to significant financial repercussions.
Initial Access and Exploitation Techniques
Threat actors have demonstrated a range of tactics to gain initial access. For instance, in mid-2026, the group known as UNC6993 compromised a cloud instance by exploiting a misconfigured Service Account key. This access enabled them to connect to a large language model (LLM) API. They also created a session hijacking page that masqueraded as an application portal, tricking users into submitting their access tokens, thereby bypassing password and multifactor authentication (MFA) controls.
In another case, a different actor combined a stolen Personal Access Token (PAT) with a suspected web application vulnerability to gain access to private code repositories. Similarly, compromised IAM credentials allowed another group to generate fresh access keys, circumventing MFA requirements. These methods illustrate the diverse strategies employed by attackers to infiltrate cloud environments.
Lateral Movement and Privilege Escalation
Once inside, threat actors engage in lateral movement to escalate privileges and solidify their access. They conduct internal reconnaissance to identify additional targets and sensitive credentials. For example, one actor performed extensive reconnaissance across multiple cloud regions, mapping available services and identity policies. In another incident, a threat actor compromised a containerized cloud environment, cloning sensitive codebases to search for hardcoded secrets and developer credentials.
Threat actors have also been observed abusing cloud identity permissions to maintain persistent access. By exploiting token-creation privileges, they generated temporary credentials for high-privilege service accounts, leading to the exfiltration of sensitive information such as internal database passwords and third-party API keys. In some cases, they created unauthorized IAM accounts with full administrative rights, injecting SSH keys into compute instances to establish backdoor access.
Monetization Strategies and Resource Hijacking
The ultimate goal of LLMJacking is to monetize compromised resources. Threat actors achieve this through various methods, including token proxying and automated model streaming. For instance, UNC6993 generated additional API keys and enabled services within victim cloud projects, resulting in over $100,000 in billing charges. They also subscribed victim accounts to premium services, employing a “burn and rotate” strategy to evade detection while maximizing their resource usage.
Moreover, some actors have leveraged their access to host proxy endpoints and agentic workloads. One group deployed a containerized backdoor in the compromised environment, enabling them to execute commands locally while querying upstream LLMs. They provisioned powerful GPU instances across multiple regions and monitored billing impacts through the cloud provider’s API.
Underground Market for Stolen Resources
Evidence suggests that threat actors are not only using hijacked resources for their operations but are also monetizing them by selling access to compromised API quotas. For example, UNC6993 utilized open-source tools to pool stolen credentials and obscure the origin of requests, likely selling access through underground storefronts. Another actor, operating the service jet-night.com, executed over 500,000 LLM API calls in a short period, passing nearly one million dollars in token costs to the victim organization. This service offered token plans significantly below market rates, indicating reliance on stolen keys for profitability.
The findings from these investigations underscore the need for organizations to enhance their security postures, particularly regarding cloud identity management and API security. As threat actors continue to evolve their tactics, understanding these vulnerabilities and implementing robust security measures will be crucial in mitigating the risks associated with LLMJacking.
For further insights into the evolving landscape of cloud security threats, refer to the detailed analysis provided by Google Cloud here.
