Fireworks AI breach (Oct 2026)
Attackers used stolen internal credentials and a breached cloud provider to reach Fireworks AI customer secrets, billing data and, for some accounts, datasets and incomplete model weights; Fireworks links them to LLMJacking.
What happened
On Oct 6 2026 Fireworks AI, an inference platform for customizing and serving open models, found that someone was using its own internal credentials. The intruder read environment variables used in evaluation jobs for a limited number of customer accounts, which captured API keys and tokens stored as secrets. Fireworks began containment the same day, started notifying affected customers on Oct 7, rotated employee API keys on Oct 9 and published a bulletin that night. At that point it said it had found no evidence of access to customer models, datasets, training data, inference traffic or personal information.
The Nov 9 update widened that account. A third-party cloud provider hosting Fireworks R&D machines had its own security incident on Oct 23; on Oct 24 the intruder used it to reach Fireworks development machines and run an unauthorized workload, which caused further customer data exposure. Fireworks says it detected that the same day and has seen no further successful access to its customer platform since Oct 25.
Data involved
According to the Nov 9 update, the affected internal application exposed:
- Resource metadata such as identifiers, display names, descriptions and creation times.
- Billing information: billing and shipping addresses, invoice and payment history, account balance and credit usage. Fireworks says it does not store payment card numbers.
- Model weights and datasets: for certain customers, a subset of incomplete model weights and datasets.
Fireworks reports no impact to Inference, the Training SDK or Nexus, which it says keep no customer data by default (zero data retention).
Downstream impact
On Nov 12 2026 TechCrunch reported that the attacker, after breaching Fireworks’ cloud provider and reaching a development system, accessed chat transcripts of some Notion customers; Notion said about 1% of its customers were affected and that it disconnected from Fireworks on Nov 3. Fireworks’ own Nov 9 bulletin lists resource metadata, billing information and, for some customers, model weights and datasets, and does not mention Notion. TechCrunch called it Fireworks’ second breach in two months.
On Nov 13 2026 Cursor said the attacker had obtained an access key Cursor issued to Fireworks, which Cursor says Fireworks exposed, and used it to copy a bucket of coding-agent traces Cursor shared for training. Credentials in those traces led into Cursor’s internal tools and, on Oct 29, to the malicious Claude Code extension in Cursor’s marketplace. Cursor moved all inference off Fireworks.
Attribution and response
Fireworks says the evidence links the attack to a financially motivated group that sells tokens from unauthorized access to large language models through illicit routing gateways, a scheme known as LLMJacking. It notified law enforcement and is working with Palo Alto Networks Unit 42 on forensics. Its listed fixes include revoking credentials and service accounts, removing unneeded elevated privileges, isolating affected systems, tightening access to data stores and improving logging. It disabled self-serve dataset uploads and downloads for Batch Inference and dataset-based training while it investigates. All of this is the company’s own account; the investigation is ongoing.
Why it matters
Inference providers hold the secrets, fine-tuning data and custom weights of many customers at once, so one compromised provider exposes all of them. The breach is part of a run of attacks on AI infrastructure; see also the GPUStack authentication bypass exploited against AI model servers.
Timeline
Every dated event in the wiki that involves Fireworks AI breach (Oct 2026), newest first.
Cursor's incident report says an attacker obtained an access key Cursor had issued to Fireworks AI, copied a bucket of mostly-2025 coding-agent traces (from sessions not in Privacy Mode), and used unredacted credentials in those traces to reach Cursor's SaaS tools. A stolen credential then let the attacker plant tampered copies of Anthropic's Claude Code extension (2.1.312 and 2.1.313) in the cache serving Cursor's marketplace on Oct 29. Cursor says under 0.05% of users had traces exposed, under 0.01% of customers had API credentials exposed, the last attacker activity was blocked Nov 1, and it moved all inference off Fireworks.
- Fireworks AI breach (Oct 2026) reached Notion customers' chat transcripts
TechCrunch reported that the attacker who breached Fireworks AI's cloud provider accessed some Notion customers' chat transcripts; Notion said about 1% of customers were affected and that it disconnected from Fireworks on Nov 3. Fireworks' latest bulletin does not mention Notion.
Fireworks AI widened its October disclosure: beyond secrets in evaluation jobs, attackers reached account and billing data and, for some accounts, datasets and incomplete model weights, partly through a breached third-party cloud provider hosting its R&D machines (Oct 23-24). It links the actor to LLMJacking; Unit 42 is assisting the investigation.
Connections
Fireworks AI breach (Oct 2026) and everything one link away. Drag to move, scroll to zoom, click a node to open it.Open in the full graph →
What links here
Company Cursor (Anysphere) · Fireworks AI · Palo Alto Networks
Incident GPUStack authentication bypass (CVE-2026-102895) · Malicious Claude Code extension in Cursor's marketplace
Concept LLMJacking
Sources
- Fireworks AI, “Fireworks October 2026 security incident”, Oct 9 2026
- Fireworks AI, “Update on October 2026 security incident”, Nov 9 2026
- TechCrunch, “Cyberattack on Fireworks AI allowed hackers to access Notion customers' chat transcripts”, Nov 12 2026
- Cursor, “An update on our recent security incident”, Nov 13 2026