Fireworks AI has confirmed that a financially motivated threat actor compromised its infrastructure. The company links the activity to LLMJacking, where stolen large language model access is resold through illicit routing gateways. Fireworks AI Reveals LLMJacking Cyberattack The attacker obtained environment variables from evaluation jobs in several customer accounts, including API keys and tokens stored as secrets.
Fireworks collaborated with Palo Alto Networks Unit 42 to uncover exposed resource metadata, billing addresses, invoice histories, account balances, and credit usage.
On October 24, the attacker exploited this breach to gain access to Fireworks development machines and ran an unauthorized workload, resulting in further customer data exposure. Mandiant and Google Threat Intelligence Group confirmed that stolen cloud identity tokens, such as IAM credentials and service account keys, were the most common entry method, with victims facing expenses up to $1 million.
FortiGuard Labs disclosed a scenario where an AWS IAM key was leaked, leading to the creation of a new IAM user and acceptance of AWS Marketplace agreements to access Amazon Bedrock models. The company has disabled self-serve dataset uploads and downloads used by Batch Inference and dataset-based training, hardening its work with credential lockdown, network isolation, tighter storage access controls, and improved logging. Partner with 16,000+ SOC teams using ANY.RUN to streamline threat investigations and reduce manual effort.












