November 10, 2026 – AI infrastructure company Fireworks AI has revealed another security breach involving a third-party cloud provider, expanding concerns about the safety of customer information and AI development systems. The company confirmed that unauthorized access affected account and billing information, while certain customer datasets and incomplete AI model files were also accessed.
According to Fireworks AI’s November 9 security update, the latest findings expand the known impact of an earlier incident. The company first detected suspicious activity on October 6, when an unauthorized party used internal credentials to access stored information in a limited number of customer accounts. Its original October security bulletin focused on exposed API keys and access tokens, with no evidence at that time that customer models, datasets or personal information had been accessed.
The investigation later identified another security incident involving an outside cloud provider that hosted Fireworks’ research and development machines. The provider experienced a breach on October 23. On October 24, an unauthorized party used that access to reach Fireworks development systems and run an unauthorized workload, causing additional customer data exposure. Fireworks says it detected the activity that day and immediately began taking action to contain it.
The affected internal application provided access to several types of customer information, including resource identifiers, names, descriptions and creation dates. Billing information such as addresses, invoices, payment history, account balances and credit usage was also involved. For certain customers, attackers accessed some datasets and incomplete AI model files. Fireworks confirmed that payment card numbers were not affected because the company does not store them.
The exposure of AI model files is particularly important for developers and businesses building their own AI applications. Model weights contain information used by AI systems to produce results, while training datasets can include material used during development. Unauthorized access could raise concerns about confidential projects or valuable research. However, Fireworks has not publicly confirmed that the affected files were published, sold or used to reproduce customer models.
Fireworks says its investigation has found no impact on its Inference, Training SDK or Nexus services, which use zero data retention by default. This means the company has not identified exposure involving those services in its investigation so far. However, it has temporarily disabled self-service dataset uploads and downloads for Batch Inference and dataset-based training while security checks continue. Its Training SDK remains available.
The company has revoked and replaced affected credentials, restricted access between systems and strengthened security monitoring. It is working with Palo Alto Networks’ Unit 42 cybersecurity team to investigate the incidents and improve its defenses. Fireworks also says it has found no further evidence of successful unauthorized access to its customer-facing platform since October 25, although the investigation remains ongoing.
Fireworks believes the activity may be linked to financially motivated cybercriminals involved in a practice called LLMJacking. This involves stealing access to AI services and using or reselling that access without permission. Separate research from Google Threat Intelligence Group and Mandiant explains how stolen cloud credentials can allow criminals to generate costly AI requests. However, Fireworks has not publicly identified a specific attacker or confirmed the full extent of any related misuse.
The company is contacting customers whose information was affected and providing instructions based on their individual accounts. Customers receiving notifications should follow the official guidance, replace affected API keys, revoke exposed third-party credentials through their original providers and review account activity for anything unusual. TechBonafide’s guide to API security risks also explains why protecting access keys, limiting permissions and monitoring connected services are important security measures.
Several important questions remain unanswered, including the total number of affected customers, the identity of the compromised cloud provider and the full extent of the exposed datasets. The incident highlights a wider concern for businesses relying on cloud-based AI tools: security problems in development systems or outside infrastructure providers can put sensitive information at risk, even when core customer-facing services show no signs of compromise. Fireworks says it will continue monitoring its systems and provide further updates as its investigation progresses.
