AI infrastructure provider Fireworks, a critical partner for startups and large enterprises developing custom artificial intelligence models, revealed on Monday that unauthorized actors accessed proprietary datasets and models over the preceding month.

The Hidden Vulnerabilities of Corporate AI Customization

This security failure highlights a growing enterprise risk as companies race to deploy cost-effective, bespoke AI systems. Organizations frequently transmit sensitive intellectual property and confidential customer data to third-party platforms like Fireworks, transforming these intermediaries into high-value targets for malicious hackers.

Fireworks serves approximately 10,000 corporate clients, featuring major brands like Uber, Shopify, Upwork, and DoorDash, alongside prominent startups such as Cursor and Cognition. These companies leverage Fireworks to train specialized AI frameworks using proprietary data atop budget-friendly, frequently Chinese-origin models.

Rapid Financial Growth and Escalating Threats

The firm generated roughly $250 million in revenue for 2025, according to previous reports. Its valuation skyrocketed fourfold to $17.5 billion within a nine-month span, propelling CEO Lin Qiao into the billionaire ranks in July with an estimated 6% stake valued at $1 billion.

Interviews with affected clients revealed significant unrest; two enterprises substantially migrated their operations away from Fireworks' ecosystem. One client characterized the security failures as a succession of escalating hacks compounded by inadequate communication from the vendor. Meanwhile, two other clients stated they remained uninformed of any security breaches.

Investigating the LLMjacking Incidents

In an official security advisory, the enterprise attributed the breach to "LLMjacking," a methodology where malicious actors hijack unauthorized access to AI models for resale, leaving victims burdened with enormous computing bills. The company confirmed it partnered with Unit 42, the threat intelligence division of Palo Alto Networks, to spearhead the forensic investigation.

Management noted that the recent intrusion expanded upon an earlier security breach disclosed in early October. An additional incident on October 24 compromised internal research and development servers along with supplementary client files. Consequently, the attackers gained access to client billing details, physical addresses, transaction histories, and fragments of unreleased AI architectures.

The Rise of Open-Source Customization

The early October incident involved compromised internal credentials affecting select accounts. CEO Qiao noted via social media that the sophisticated behavior of the attackers suggested the involvement of automated AI agents.

Fireworks' meteoric ascent coincided with surging operational costs for AI deployment. Rather than relying exclusively on expensive commercial models from established American laboratories, developers increasingly turn to open-source models for greater financial and operational control. Companies like Coinbase slashed their AI expenditures by half by directing routine coding tasks to affordable Chinese models, while DoorDash delegates preliminary code reviews to platforms like Moonshot's Kimi K2.6 while retaining premium frontier models for complex logic.