AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: What The Fireworks Breach Revealed About Notion AI Chat Data on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the little things that make your day delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

Notion says a dataset accessed in the Fireworks AI breach included stored Notion AI chat transcripts tied to its Insights feature. The company puts the affected share below 1% of workspaces and says it revoked exposed API keys found in affected messages on November 11. Notion has not said which Fireworks system retained the data or how long it was stored.

Notion says a dataset accessed during the Fireworks AI breach included stored Notion AI chat transcripts used for its Insights feature, and that Fireworks retained data despite a contractual commitment to zero data retention. Notion says fewer than 1% of workspaces were affected and that it revoked exposed Notion API keys found in affected user messages on November 11.

In its updated security advisory and Help Center notice, Notion said the dataset Fireworks provided on November 8 included transcripts from Notion AI chats used for Insights, a feature that gives workspace administrators information about how teams use AI. Notion said that, for the vast majority of affected workspaces, the transcripts were unattributable: they were not labeled with the author or workspace and did not contain tool-call results.

That description does not mean the transcript contents were necessarily harmless. Notion defines a transcript as any input to or output from the model written to the chat. It advises affected customers to review the data because a message may contain sensitive information. On November 11, the company also said it revoked exposed Notion API keys found in affected user messages.

Notion said AI Meeting Notes—including audio, transcripts and summaries—were not affected unless they had been used as an input in a Notion AI chat transcript. It said no other Notion features were affected. The company advises affected customers to treat the data as exposed, review it under its Security incident review guidance, and rotate passwords, API keys or other credentials entered into Notion AI during the incident.

At a glance
updateWhen: Notion updated its security notice on N…
The developmentNotion expanded its Fireworks breach notice to say the accessed dataset included Notion AI chat transcripts and that data had been retained despite Fireworks’ contractual zero-retention commitment.
AI data security · Briefing

What the Fireworks Breach Revealed About Notion AI Chat Data

Notion says a dataset accessed in the Fireworks AI breach included stored chat transcripts used for its Insights feature. The disclosure raises questions about data retention, exposure scope and vendor commitments.

<1%
of workspaces affected
November 11
exposed API keys revoked

What data was involved

Notion says the accessed dataset included stored requests and responses from AI chats used for Insights.

Transcript

Chat inputs and outputs

A transcript can contain any model input or output written to the chat, including sensitive work material.

Attribution

Usually unlabeled

Notion says most affected transcripts were not labeled with an author or workspace and lacked tool-call results.

How the data reached Fireworks

Notion says it used Fireworks for Insights classification and other inference; the exposed data came from Insights.

1
Data accessed

On November 2, Fireworks informed Notion that malicious actors had accessed a small set of data.

2
Connection ended

On November 3, Notion says it disconnected from Fireworks, deleted its data there and halted training runs.

3
Notice updated

Notion updated its notice on November 10 and 11 after learning the dataset included chat transcripts.

What remains unclear

The published accounts leave important details about retention and scope unanswered.

  • Which Fireworks system retained the data and for how long.
  • How many individual users were affected.
  • Why retention occurred despite a contractual zero-retention commitment.
  • Whether other Fireworks customers faced the same issue.
  • What customers should review

    Notion advises affected customers to treat the data as exposed.

    • Review relevant chat inputs and outputs for sensitive information.
    • Rotate passwords, API keys or other credentials entered into Notion AI during the incident.
    • Follow Notion’s Security incident review guidance.
    Scope noteNotion says AI Meeting Notes were not affected unless used as input in a Notion AI chat transcript; no other Notion features were affected.
    The takeaway
    The breach exposed stored Insights chat transcripts, making both vendor data-retention practices and targeted customer review central to understanding the risk.

    What the Retention Failure Means

    The disclosure matters because it concerns both what data was accessed and why the data was present in Fireworks’ systems. Notion said Fireworks had made a contractual commitment to zero data retention for Notion’s use of its platform, but that its investigation found data retained from the endpoints Notion used. That is a gap between the stated commitment and the handling of data described in the company’s notice.

    Notion also says its own zero-retention commitments apply to live API requests sent to model vendors to generate responses. Stored information used to provide features such as chat history and Insights is outside that scope, according to its FAQ. In this case, Notion says stored requests and responses were sent to Fireworks for Insights analysis. For organizations evaluating AI services, the distinction shows that a promise about live inference may not describe how data is handled for supporting or analytics features.

    The transcript detail also makes the potential exposure relevant to users who treated chat as a place to enter work material. Notion recommends reviewing affected data for sensitive content; it has not said that every transcript contained such information. The reported exposure therefore calls for a targeted review, not an assumption that all affected chats included credentials or confidential records.

    Products Worth Considering

    LEVOIT Vital 200S-P Smart Air Purifier for Large Rooms up to 1875 ft², Quiet 27 dB Operation, AHAM Verifide, HEPA Sleep Mode, White

    LEVOIT Vital 200S-P Smart Air Purifier for Large Rooms up to 1875 ft², Quiet 27 dB Operation, AHAM Verifide, HEPA Sleep Mode, White

    As an affiliate, we earn on qualifying purchases.

    How Notion’s Fireworks Use Worked

    Notion says it used Fireworks for two functions: classification jobs for the Insights feature and as a general inference provider for open-weight models within Notion. According to the updated notice, the exposed data came from the Insights side. The company says those jobs involved stored requests and responses from Notion AI use, rather than only live prompts being processed to generate an answer.

    Notion’s published timeline says Fireworks informed it on November 2 that a small set of data had been accessed by malicious actors. Notion initially understood the material to be synthetic testing data. On November 3, it says it disconnected its systems from Fireworks, deleted its data from Fireworks’ system and halted training runs. Notion later updated its notice on November 10 and 11 after learning the dataset included chat transcripts. Fireworks’ own bulletin, according to the source material, still showed a last update date of November 9.

    Fireworks has said the incident involved access to data, including datasets, for certain customers. Its update also said it had not detected an impact to Inference, Training SDK and Nexus, which it described as zero-data-retention by default. Notion’s account does not identify the Fireworks system involved in retaining its data, so the two companies’ published descriptions do not establish how the retained transcripts fit into Fireworks’ broader account of the incident.

    “Unfortunately, during our investigation we learned that data was retained from the Fireworks endpoints that we used.”

    — Notion, in its updated Help Center notice

    Questions About Retention and Scope

    Notion has not identified which Fireworks system retained the data, how long it was kept or how many individual users were affected. The company gives the share of affected workspaces as fewer than 1%, but the notice does not provide a total number of workspaces or a count of people. It also remains unclear whether other Fireworks customers encountered the same retention issue.

    The published accounts do not explain why data was retained despite the contractual commitment or how that retention relates to Fireworks’ statement that certain products are zero-retention by default. The available material does not establish that those statements refer to the same system or data flow. Notion has said it is contacting affected workspace owners; according to its notice, if neither a user nor the workspace owner has heard from the company, Notion has not identified that workspace as affected.

    Customer Reviews and Vendor Audit

    Notion says it is auditing critical vendors, including inference providers and data processors, for data-minimization practices and zero-retention policies. Affected workspace owners should follow the company’s incident-review guidance, examine relevant chat inputs and outputs, and rotate credentials that may have been entered during the incident. Notion says it has revoked exposed API keys found in affected messages, but customers should still review any other credentials in the data.

    Further clarity depends on updates from Notion and Fireworks about the retained data, the system involved and the scope across customers. Notion’s notice says it will contact affected workspace owners; the supplied information does not give a date for another update. Until the companies publish more detail, the duration of retention and the number of people whose messages may have been included remain unknown.

    Key Questions

    What Notion AI data was included in the dataset?

    Notion says it included chat transcripts used for the Insights feature. A transcript can contain model inputs and outputs written in the chat, though Notion said most affected workspaces’ transcripts were not labeled with a user or workspace.

    How many Notion workspaces were affected?

    Notion says fewer than 1% of workspaces were affected. It has not published an exact workspace count or the number of individual users involved.

    What should affected users do?

    Notion advises affected customers to treat the data as exposed, review it using the company’s Security incident review guidance, and rotate passwords, API keys or other credentials entered into Notion AI during the incident.

    Were Notion AI Meeting Notes affected?

    Notion says AI Meeting Notes, including audio, transcripts and summaries, were not affected unless they were used as an input in a Notion AI chat transcript. It said no other Notion features were affected.

    Why is the zero-retention commitment in question?

    Notion says Fireworks had a contractual zero-retention commitment for its use of the platform, but Notion’s investigation found that data was retained from the endpoints it used. The companies have not said which system stored it or how long it remained there.

    Source: ThorstenMeyerAI.com

    You May Also Like

    Reading OpenAI’s EU Agentic AI Claim: Start With The Denominator

    OpenAI says agents generated 55% of output tokens in its EU business accounts, but the figure does not measure tasks, users or business value.

    Could Automation Reduce Manual Checklist Work For Brokers?

    A proposed brokerage workflow would turn new CRM deals into transaction checklists and draft e-signature packets, with human approval before sending.

    A War Room for Your Next Idea: Inside IdeaClyst

    Discover how IdeaClyst provides founders with a local AI-driven war room to validate ideas, reduce risk, and make strategic decisions efficiently.

    Why Anthropic Is Restricting Claude Code’s Game Recreation

    A GamesIndustry.biz headline reports a new Claude Code rule targeting recreations of existing games, but its scope and enforcement are unknown.