Fireworks AI — AI Infrastructure Breach Exposes Uber Data via Supplier Credentials
Publication date: November 17, 2026
Category: Supply chain attacks (realtime trigger)
Introduction
A major security breach at artificial intelligence infrastructure platform Fireworks AI has compromised personal data belonging to nearly 150,000 Uber users and drivers. The incident unfolded when malicious actors utilized stolen internal credentials belonging to a technology provider to gain unauthorized access to connected corporate environments. Subsequent investigations revealed that the campaign—linked to monetization objectives through AI resource abuse known as LLMjacking—also impacted other prominent tech companies such as Notion and Cursor, exposing systemic vulnerabilities within AI supply chains.
What is Fireworks AI and the AI Supply Chain Risk? (General Analysis)
Fireworks AI is a cloud platform that enables developers to host, run, and fine-tune large language models (LLMs) and artificial intelligence infrastructure without managing the underlying hardware directly. To fulfill these capabilities, such platforms handle high levels of administrative privileges, corporate secrets, API access keys, and confidential datasets.
When a provider of this nature suffers a breach, customer organizations relying on their services inherit the risk indirectly. In this incident, the vector was not a software bug or vulnerability, but a failure in Identity and Access Management (IAM) controls at the supplier level.
- Estimated CWE Classification: CWE-284 (Improper Access Control) and CWE-522 (Insufficiently Protected Credentials).
- Estimated Severity: Critical (due to the massive scope of exposed identity data and the loss of control over execution environments).
How Does It Work? (Technical Analysis)
Forensic analysis and reports from incident response teams (including Palo Alto Networks Unit 42) mapped out the methodology employed by the threat actors:
- Initial Access Vector: Attackers acquired legitimate internal credentials belonging to Fireworks AI. The exact origin points toward token-theft campaigns or social engineering directed at supplier personnel or systems.
- Security Control Evasion: Because the attackers operated using valid and authorized credentials within the Fireworks ecosystem, Uber’s monitoring systems and configured alerts failed to detect any anomaly. Since the requests appeared completely legitimate from an authentication standpoint, conventional security mechanisms triggered zero alarms (“no alerts were triggered”).
- Access and Exfiltration: Once inside the external environment linked to Uber, the malicious actors accessed stored records containing sensitive information, including names, home addresses, dates of birth, and images of driver’s licenses and government-issued IDs. Subsequent activities also tied into LLMjacking schemes, where criminals utilize stolen tokens to execute models at the victims’ expense.
Affected Systems / Environments
The incident primarily affected cloud environments managed by Fireworks AI and its clients’ corporate integrations. Impacted systems and data include:
- Fireworks AI Platform: Customer accounts, billing details, corporate secrets, training datasets, and unfinished AI models.
- Uber Data: Personally Identifiable Information (PII) of approximately 149,346 individuals (primarily drivers), including driver’s license images.
- Other Affected Providers/Clients: Notion and Cursor confirmed similar exposures stemming from the same supplier supply chain incident.
Mitigation and Detection
Remediation
To counter supply chain threats of this nature, organizations and service providers must enforce rigorous hardening measures:
- Immediate Rotation and Revocation: Instantly replace all compromised master credentials, API tokens, and stored secrets.
- Principle of Least Privilege: Restrict privileged access within AI platforms strictly to essential personnel, properly segmenting development and production environments.
- Identity Behavior Monitoring: Implement User and Entity Behavior Analytics (UEBA) to detect anomalous usage patterns even when utilized credentials are technically valid (e.g., access from unexpected geographic locations or atypical model invocation volumes).
Detection
Defensive teams (Blue Teams) must closely monitor interactions with third-party services and AI APIs:
“The use of legitimate credentials by threat actors demonstrates that traditional authentication relying on static secrets is insufficient against supply chain breaches; deep visibility into connected account behavior is required.”
- Conceptual Indicators of Compromise (IoCs): AI API call patterns outside standard working hours, sudden spikes in compute consumption for model inference, and access requests targeting storage buckets or external databases originating from IPs tied to commercial VPNs or anonymous networks.
- Audit Rules: Configure strict alerts in IAM systems for permission modifications on third-party service accounts and mass authentication token generation.
Wrapping Up
The security breach at Fireworks AI underscores the inherent risks associated with relying on third-party AI infrastructure providers. By compromising the supplier’s internal credentials, attackers bypassed Uber’s traditional detection systems and exfiltrated highly sensitive data belonging to thousands of drivers. This case highlights the urgent need to reassess implicit trust granted to external integrations and adopt Zero Trust architectures paired with advanced identity behavior monitoring.
References
- Security Affairs. (2026, November 17). Fireworks AI Breach Exposes Uber Data and Highlights Supply Chain Risks. Retrieved from https://securityaffairs.com/?p=203339
- TechCrunch. (2026). Reports regarding the security breach at Fireworks AI, Uber, Notion, and Cursor.
- Reuters. (2026). Reports on data breach notifications and investigations involving Uber.
- Palo Alto Networks Unit 42. (2026). Incident response updates regarding the Fireworks AI security incident.