Skip to content

Fireworks AI breach leaks Notion logs: supply-chain gaps expose conversation texts and billing data

Share
Fireworks AI breach leaks Notion logs: supply-chain gaps expose conversation texts and billing data

Listen to this article

Read by Anchor

When a software company embeds an external artificial-intelligence interface into the core of its tools, it opens a security window it cannot fully close. This lesson has returned to affect collaborative workspaces after the Fireworks AI platform, which provides open models to developers, disclosed a breach that compromised its cloud-service provider. The intrusion reached an internal development system and from there accessed some customers’ data, billing information and technical datasets, but the fastest fallout hit users of the document-management and note-taking platform Notion directly.

Notion, which relies on Fireworks to run part of its AI features, announced that the attack exposed logs of user conversations that used the analytics feature in its generative tool. The breach affected roughly one percent of the platform’s user base, including full conversation texts with both user-entered inputs and model-generated outputs. Although the company indicated that a large portion of the leaked data was not directly linked to customer identities, theRecording of generative dialogues revealed work drafts and sensitive notesthat teams deposit in daily editing spaces without accounting for cloud-crossing risks.

Notion’s management quickly disconnected all its systems from Fireworks on 3 November, erased stored data and cut every technical communication line between the parties. This step exposes the intertwined technical dependence between software developers and inference intermediaries. Targeting major platforms no longer requires breaching their primary defenses; instead, the intermediate infrastructure and hosting providers have become the preferred attack vector, allowing a breach to flow from inference servers to contracted companies’ repositories in moments.

The attackers’ motives reflect a notable economic shift in hacking patterns aimed at the AI sector. Fireworks, which serves tens of thousands of global clients including companies such as Harvey and Cipher, explained that the perpetrators were financially motivated, focusing on stealing inference quotas and reselling digital tokens through unauthorized access to large language models. This illicit trade lets third parties burn cloud-processing capacity without paying for it, and it marks the second incident to hit the platform within two consecutive months after a similar security event in October.

This incident pushes CTOs and information-security leaders in Gulf and Egyptian banks and firms toward mandatory supply-chain audits.Relying on a certified software platform no longer guarantees data safetyIf its generative features depend on intermediary inference providers to process prompts, the leakage of editing notes and strategy drafts raises the practical question of whether sending sensitive contexts to external cloud services is worthwhile, and it calls for a review of internal access permissions, verification of sub-model provider chains, or a shift toward hosting models in isolated environments that prevent operational commands from leaking to uncontrolled intermediaries.

Don't miss the next story

Subscribe for updates