13 November 2026 — Fireworks AI, a platform developers use to run and train open AI models, has disclosed that an October security incident reached more customer data than it first reported, including billing information and, for some customers, datasets and incomplete model weights. Productivity software company Notion has separately confirmed that data taken in the incident included Notion AI chat transcripts from fewer than 1% of its workspaces.
In an update to its security bulletin, last revised on 9 November, Fireworks said that when it first published the bulletin on 9 October, only secrets stored by Fireworks, such as those used in evaluation jobs, were known to have been accessed. "Since then, we discovered impacts to account and billing information. For a limited set of accounts, we also discovered impact to secrets, datasets, and incomplete models," the company said.
How the intrusion widened
According to Fireworks, a third-party cloud provider that hosted its research and development machines suffered a security incident on 23 October. On 24 October, the unauthorised party used that access to reach Fireworks development machines and run an unauthorised workload that led to further customer data impact. Fireworks says it detected the activity the same day and that, as of 25 October, it had not found further evidence of successful unauthorised access to the platform its customers use.
The company says the affected internal application gave access to resource metadata such as identifiers and display names; billing information including billing and shipping addresses, invoice and payment history, account balance and credit usage; and, for certain customers, a subset of incomplete model weights and datasets.
“Payment card numbers are not stored by Fireworks and were not impacted.”
Fireworks AI, Update on October 2026 security incident
Fireworks says it has found no impact to its Inference, Training SDK and Nexus products, which it describes as zero data retention by default. It has notified law enforcement, is working with Palo Alto Networks Unit 42 on forensic analysis, and has disabled self-serve dataset uploads and downloads used by Batch Inference and dataset-based training while it investigates. It also says the evidence points to a financially motivated group that sells tokens obtained through unauthorised access to large language models, a practice it calls "LLMJacking".
Notion's disclosure
Notion's advisory sets out a timeline. On the evening of 2 November, Fireworks told Notion that a small set of data had been accessed; Notion says that dataset was synthetic testing data with no customer data. Notion disconnected its systems from Fireworks by the morning of 3 November and deleted all of its data in the Fireworks system on 4 November. On 6 November, Fireworks told Notion that potential customer data may have been accessed, and it confirmed the final affected dataset on 8 November.
“Upon analysis of the dataset Fireworks AI gave us on November 8, we discovered that it included Notion AI chat transcripts used for the Insights feature, which for the vast majority of impacted workspaces were unattributable.”
Notion, Fireworks AI incident advisory
Notion explains that "unattributable" means the chats were not labelled with who wrote them or which workspace they came from, and did not contain tool call results. It says no other Notion features were affected, that it contacted affected customers on 9 November, that it has disabled the Insights feature, and that it no longer serves open-weight models through Fireworks. In later updates, Notion says it revoked exposed Notion API keys found in affected user messages on 11 November and, on 12 November, reported exposed third-party API keys found in those messages to the vendors that issued them.
Notion recommends that affected customers treat the data as exposed and rotate any passwords, API keys or other credentials they pasted into Notion AI during the incident period.
Background
Notion says it used Fireworks for classification jobs behind Insights, a feature that helps customers understand their own AI usage, and as a general inference provider for open-weight models in its product. Fireworks describes the incident as connected to a wider rise in LLMJacking and refers readers to a 2 November blog post on the subject by Google Threat Intelligence Group and Mandiant. Both companies say their investigations are ongoing.
“Even though this was not a breach of our systems, keeping your data safe is fundamental to using Notion for your life’s work.”
Notion, Fireworks AI incident advisory
Analysis
In my view, this incident shows how AI features quietly extend a company's supply chain. A Notion user typing into an AI chat would reasonably think of that text as staying with Notion, yet it travelled to an inference provider, and that provider's own cloud provider became the way in. Notion's decision to revoke and report API keys it found inside chat messages is a useful reminder that people paste secrets into AI tools far more often than they should.
If your team uses AI assistants inside SaaS products, it is worth checking which sub-processors handle those prompts, keeping secrets out of chats entirely, and rotating anything that was pasted in. The staged disclosure here, from "synthetic test data" to customer transcripts within a week, is also a reason to treat early incident statements as provisional until the final scope is confirmed.
About this report
Based on Fireworks AI's security bulletin update of 9 November 2026 and Notion's security advisory, originally published on 12 November 2026 and updated on 13 November 2026. Facts and quotes are taken from the sources credited below, which I read on 13 November 2026, and checked against the further reporting listed. Sections marked "Analysis" are my own view. I am an independent developer and am not affiliated with the organisations named here.
Credits and sources
- Source: Fireworks AI: Update on October 2026 security incident (last updated 9 November 2026)
- Source: Notion: Fireworks AI incident security advisory (published 12 November 2026, updated 13 November 2026)
- Further reporting: Zack Whittaker, TechCrunch, 12 Nov 2026: Cyberattack on Fireworks AI allowed hackers to access Notion customers' chat transcripts
- Further reporting: Bill Toulas, BleepingComputer, 11 Nov 2026: Notion says Fireworks breach exposed customer AI chat transcripts
- Lead image: Illustration created for this article: computer-generated graphic with headline text, made by Depakar B.
Corrections
Spotted an error? Please contact me. Corrections are noted on this page with the date.