Fireworks AI Data Breach Widens After Third-Party Cloud Provider Incident
Fireworks AI has disclosed a more extensive data breach, with attackers gaining access to customer data and incomplete AI models through a compromised third-party cloud provider.
Fireworks AI has revealed a significant expansion of a previously disclosed data breach, now confirming that attackers accessed customer account details, billing records, datasets, and even incomplete AI model weights. The breach, initially believed to be limited to exposed credentials, has been found to have a much wider impact due to a security incident at a third-party cloud provider that hosts Fireworks' research and development infrastructure.
The expanded scope of the breach came to light in an updated disclosure on November 9, 2026. The initial incident, detected on October 23, involved a security lapse at a cloud provider, which attackers exploited the following day to gain access to Fireworks' development machines. This unauthorized access led to the exposure of additional sensitive customer information, a stark escalation from the earlier report of unauthorized access to secrets used in evaluation jobs across a limited number of accounts.
Fireworks AI's investigation confirmed that the attackers accessed resource identifiers, names, descriptions, and creation times. Billing information, including addresses and payment history, was also compromised, though the company stated that payment card numbers were not stored and therefore unaffected. Crucially, for some customers, the attackers were able to access subsets of unfinished AI model weights and datasets, although complete production models were reportedly not stolen.
The company has not identified the specific cloud provider or the vulnerabilities exploited. However, they confirmed detecting the malicious activity on October 24 and immediately implemented response measures, including closing identified critical vulnerabilities. As of October 25, no further successful access by the attacker on Fireworks' customer platform was detected.
This incident is believed to be the work of financially motivated cybercriminal groups aiming to profit from large language models, potentially through illicit resale channels, a practice known as LLMJacking. Cybersecurity News has previously reported on similar attacks targeting AI services and the exploitation of cloud credentials for financial gain.
In response to the breach, Fireworks AI has taken several steps, including notifying law enforcement and engaging Palo Alto Networks Unit 42 for forensic analysis. The company is also implementing enhanced security measures such as replacing credentials, reducing elevated permissions, isolating affected systems, and tightening storage access controls. For affected customers, Fireworks has advised rotating exposed third-party credentials, reviewing access logs for suspicious activity, and replacing revoked Fireworks keys.
To further mitigate risks during the ongoing investigation, Fireworks has temporarily disabled self-service dataset uploads and downloads for specific services. While Training SDK and Fireworks Lab-based training remain available, they are being run in isolated environments with enhanced security. The company is directly contacting all customers whose data may have been affected by this incident.
The breach underscores the growing risks associated with the AI supply chain and the critical importance of robust security practices for third-party vendors. As AI models become more valuable, they also become more attractive targets for cybercriminals seeking to exploit them for profit.