Fireworks AI Confirms Wider Breach Affecting Customer Datasets, Model Weights and Notion AI Chats

Fireworks AI has disclosed a significant expansion of a security breach initially reported last month, now affecting customer datasets, model weights, and Notion AI chats. The AI infrastructure provider revealed that the breach involved unauthorized access to its development environment through a third-party cloud provider, impacting a limited set of accounts. This breach highlights the vulnerabilities in AI systems, where unauthorized access can lead to exposure of sensitive data, including account and billing information, secrets, and incomplete models.

The breach was first detected on October 6, 2026, when Fireworks AI identified unauthorized use of internal credentials. Initially, it was believed that only environment variables, such as API keys and tokens, were accessed. However, further investigation revealed that attackers exploited the development environment on October 23, 2026, to run unauthorized workloads, leading to broader data exposure. The company has since revoked compromised credentials and rotated service account tokens to mitigate the impact.

Fireworks AI's latest update indicates that attackers accessed its development machines hosted by an unnamed third-party cloud provider. This access allowed them to execute unauthorized workloads and potentially impact additional customer data. The breach is linked to a financially motivated cybercriminal group involved in LLMjacking, a threat that hijacks large language models using stolen credentials. LLMjacking involves selling unauthorized access to these models, leaving legitimate owners to bear the costs.

The breach has affected fewer than 1% of Notion workspaces, with potential access to Notion AI chat transcripts used for the Insights feature. Notion has confirmed that these transcripts were mostly unattributable, meaning they lacked identifiable information about their authors or workspaces. Despite this, Notion has disabled the Insights feature and stopped serving open-weight models through Fireworks while continuing its investigation. Customers are advised to review impacted data and rotate any sensitive information or credentials shared during the attack window.

This incident underscores the risks associated with AI infrastructure, where compromised credentials can lead to significant data exposure. Attackers often target cloud environments due to their centralized nature and potential for accessing vast amounts of data.

Fireworks AI has implemented several security hardening procedures to prevent further unauthorized activity, including isolating affected systems, tightening access controls, and enhancing logging and monitoring systems. The company is also alerting affected customers and working to secure its systems against future attacks. As AI continues to play a crucial role across various sectors, including financial services and technology, ensuring the security of AI infrastructure is paramount to protect sensitive data and maintain trust.