Cybercriminals Hijack Enterprise AI Services, Driving Six-Figure Cloud Charges in Hours

 

LMJacking Attacks Are Driving Massive Cloud Charges

Security researchers are warning that LLMjacking has become a rapidly growing threat, with some organizations accumulating more than $100,000 in fraudulent AI usage charges in a single day and total losses reaching $1 million before the abuse was detected and stopped.

What is LLMjacking?

LLMjacking occurs when attackers steal or abuse access to an organization's AI infrastructure, typically by obtaining:

• LLM API keys
• Cloud IAM credentials
• Service account keys
• GitHub Personal Access Tokens (PATs)
• Active cloud session tokens

The attackers then use the victim's AI resources to run their own workloads while the victim pays the bill.

Why it's worse than cryptojacking

Unlike cryptojacking, which generally consumes resources gradually, LLMjacking can generate huge cost spikes within hours because modern AI models are expensive to operate. Researchers observed cases where unauthorized AI requests surged to thousands of requests per second, generating six-figure costs almost immediately.

How criminals profit

Google's Threat Intelligence Group has identified a growing criminal ecosystem that includes:

1. Initial access brokers who steal credentials.
2. LLM Mixers that aggregate stolen AI access.
3. Reseller services that offer customers cheap or "unlimited" access to premium AI models using compromised accounts.

In one case, a service known as Jet Night allegedly routed more than 500,000 API calls through stolen credentials, passing nearly $1 million in AI costs to the victim organization.

Common attack methods

Attackers are gaining access through:

• Exposed API keys in source code
• Phishing campaigns
• Compromised cloud credentials
• Misconfigured service accounts
• Vulnerable AI gateway software
• Session token theft and session hijacking attacks

Mitigation Recommendations

For enterprises and financial institutions:

• Store API keys in a secrets manager, never in code.
• Enable MFA and use phishing-resistant authentication.
• Rotate AI and cloud credentials regularly.
• Monitor AI service consumption for sudden spikes.
• Set billing alerts and hard spending limits.
• Restrict AI API access by IP address where possible.
• Continuously scan repositories for exposed secrets.
• Audit service accounts and unused permissions.