---
type: "post"
id: "update-on-october-2026-security-incident-c8af72"
slug: "update-on-october-2026-security-incident-c8af72"
title: "Update on October 2026 security incident"
description: "Fireworks AI's November 9 update says an October incident involving a third-party cloud provider extended into its development environment and affected customer billing data, with secrets, datasets and incomplete model weights exposed for a limited set of accounts."
retrieval_nugget: "Fireworks AI's November 9 update says an October incident involving a third-party cloud provider extended into its development environment and affected customer billing data, with secrets, datasets and incomplete model weights exposed for a limited set of accounts. The company says its zero-data-retention services were not affected, while some dataset workflows remain disabled during its investigation."
published_at: "2026-11-10"
updated_at: "2026-11-10"
record_date: "2026-11-10"
date_kind: "published_at"
topics: ["inference"]
entities: ["Fireworks AI","Palo Alto Networks Unit 42","Inference","Training SDK","Nexus","Batch Inference"]
related_patterns: ["agent-security-moves-to-runtime-boundaries"]
novelty_label: "new_fact"
what_changed: "Fireworks AI disclosed that its October cloud-provider breach also affected account and billing information and, for a limited set of accounts, secrets, datasets and incomplete models."
prior_refs: []
source_url: "https://fireworks.ai/security/bulletins/october-2026-security-incident-update"
source_urls: ["https://fireworks.ai/security/bulletins/october-2026-security-incident-update"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:update-on-october-2026-security-incident-c8af72"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/update-on-october-2026-security-incident-c8af72/","markdown":"https://newruntime.com/posts/update-on-october-2026-security-incident-c8af72.md","json":"https://newruntime.com/posts/update-on-october-2026-security-incident-c8af72.json"}
markdown_url: "https://newruntime.com/posts/update-on-october-2026-security-incident-c8af72.md"
---

# Update on October 2026 security incident

## Retrieval answer

Fireworks AI's November 9 update says an October incident involving a third-party cloud provider extended into its development environment and affected customer billing data, with secrets, datasets and incomplete model weights exposed for a limited set of accounts. The company says its zero-data-retention services were not affected, while some dataset workflows remain disabled during its investigation.

## Summary
Fireworks AI's November 9 update says an October incident involving a third-party cloud provider extended into its development environment and affected customer billing data, with secrets, datasets and incomplete model weights exposed for a limited set of accounts. The company says its zero-data-retention services were not affected, while some dataset workflows remain disabled during its investigation.

## What happened
Fireworks says this update revises a security bulletin it published on October 9. It notified law enforcement and brought in Palo Alto Networks Unit 42 to conduct forensic analysis. The company says its initial understanding was that only Fireworks-stored secrets, such as those used in evaluation jobs, had been accessed. Subsequent investigation found impacts to account and billing information, and, for a limited set of accounts, secrets, datasets and incomplete models.

The update describes a third-party cloud provider hosting Fireworks research-and-development machines as experiencing a security incident on October 23. On October 24, an unauthorized party used that access to reach Fireworks development machines and run an unauthorized workload, leading to additional customer-data impact. Fireworks says it detected the activity that day, took remediation measures and closed known critical vulnerabilities. It reports no further evidence, as of October 25, of successful unauthorized access to the Fireworks platform used by customers.

The affected internal application exposed resource metadata, including resource identifiers, display names, descriptions and creation times, as well as billing information such as billing and shipping addresses, invoice and payment history, account balance and credit usage. Fireworks says it does not store payment-card numbers and those were not impacted. For certain customers, a subset of incomplete model weights and datasets was accessed. The company says it is notifying affected customers with specific impact details and recommended response actions.

The bulletin identifies the affected categories, rather than presenting one undifferentiated breach scope. Resource metadata can establish which assets existed and when they were created; billing records can expose commercial and contact details; and secrets or training artifacts may require a different response from the affected customer. Fireworks does not say that every account or every product workflow was exposed. Its stated scope is billing and account information generally through the affected internal application, with secrets, datasets and incomplete model weights limited to certain customers.

Fireworks says it has revoked and replaced credentials and service accounts, removed unnecessary elevated privileges, restricted unauthorized network pathways, isolated affected systems, tightened data-storage and analytics access, and enhanced logging and monitoring. It has disabled self-serve dataset uploads and downloads used by Batch Inference and dataset-based training while investigating. The Training SDK remains available with zero data retention, and Fireworks Lab-based training runs in an isolated environment with additional network and storage protections. The company says Inference, Training SDK and Nexus are zero-data-retention by default and it has not detected an impact to those services.

Fireworks links the activity to a financially motivated cybercriminal group whose goal was selling tokens obtained through unauthorized access to large language models to illicit routing gateways, a practice it calls LLMJacking. That attribution is the company's assessment based on its investigation. The immediate operational change stated in the bulletin is narrower: self-serve dataset transfer for Batch Inference and dataset-based training is unavailable, while the investigation and security hardening continue.

## What the archive adds
New Runtime's earlier coverage on October 24 said OpenCode reported unauthorized access affecting 1.4% of Console users and potentially exposing linked provider API keys ([earlier note](https://newruntime.com/posts/isec-news-opencode-says-breach-tied-to-distillation-e57018/)). A November 4 note covered OpenCode's account of the same incident ([earlier coverage](https://newruntime.com/posts/opencode-breach-what-happened-and-how-to-rotate-b89dc6/)). Those are separate incidents, not prior coverage of Fireworks; this update opens a new provider-security thread.

## Trend signal
This supports the security concern behind New Runtime's runtime-boundary pattern: agent and model services depend on controls over credentials, data access and execution environments, not just prompts. The incident does not itself demonstrate a particular runtime design, so the connection is limited rather than proof of the pattern. [Agent security moves to runtime boundaries](https://newruntime.com/patterns/agent-security-moves-to-runtime-boundaries/).

## Why now
The November 9 bulletin is a material update to Fireworks' October 9 notice: the company says its investigation identified additional customer-data categories beyond the stored secrets it initially knew had been accessed. It also dates the additional cloud-provider incident to October 23 and the subsequent access to Fireworks development machines to October 24.

## Why it matters
For teams using a model platform, a provider-side development environment can become a path to customer artifacts even when the inference product itself uses zero data retention. That makes service-level labels insufficient as a complete risk assessment: the relevant questions include where datasets and checkpoints are stored, which workflows create credentials, and how quickly customers learn whether their own artifacts were involved.

Fireworks says affected customers are being notified, but the incident also creates a near-term operational issue for teams that rely on dataset uploads or dataset-based training: those self-service workflows are unavailable while the investigation continues. A team may need to separate inference workloads from training or batch workflows in its continuity planning, because the bulletin makes different statements about those surfaces.

Teams should map their Fireworks use to the named categories, review any direct notification, and avoid assuming that an unaffected inference endpoint means every connected workflow or stored artifact is unaffected. That is especially relevant where evaluation jobs, uploaded datasets, or incomplete weights are part of a development pipeline. The stated controls—credential replacement, reduced privileges, network isolation, and tighter storage access—are remediation steps, not a customer-specific confirmation of scope.

## Limits and uncertainty
Fireworks says its investigation is ongoing and may reveal more information. The bulletin does not quantify the number of affected accounts or specify how many customers had secrets, datasets or incomplete weights accessed. It also does not describe the content or sensitivity of each affected dataset or model artifact, the identity of the cloud provider, or the duration of the unauthorized access.

The company says it has not detected impact to Inference, Training SDK and Nexus, but those findings remain based on its investigation to date and have not been independently verified here. Fireworks also does not state a date for restoring self-serve dataset uploads and downloads, or whether customers should take a uniform rotation action beyond the individualized response guidance it says it is providing to impacted customers.

## Reader takeaway
Treat this as a provider security incident with distinct scopes: Fireworks says billing data was affected, a limited set of accounts also had secrets or model-training artifacts exposed, and its ZDR services showed no detected impact. Check direct notifications, map your use to the named workflows, and plan around the continuing dataset-transfer restriction rather than treating the incident as a generic inference outage.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Inference](https://newruntime.com/topics/inference.md): Explore the inference topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)