Notion says Fireworks kept AI chats despite promising zero data retention

✕Sat 14 Nov ‹›

Notion security · Sat 14 Nov

Notion says Fireworks kept AI chats despite promising zero data retention

<1%of Notion workspaces affected, Notion says

Notion says its former AI vendor Fireworks had promised in its contract not to keep Notion's data, but did. The explanation, added to Notion's help page and reported by ITmedia AI+ on 13 November, covers chats from the caught up in the Fireworks incident.

Key points

Notion's Enterprise (ZDR) promise covers the requests it sends to outside AI model vendors to answer prompts. It never covered features that must store chats, such as chat history and Insights, the admin tool that shows how members use AI.
For Insights, Notion sent stored chats to Fireworks for analysis. Fireworks gave a contractual zero-retention promise, but Notion's investigation found data was kept on the Fireworks endpoints it used. Notion disconnected Fireworks on 3 November and says all data was deleted from Fireworks' systems on 4 November.
Notion is now auditing critical vendors, such as providers and data processors, to check they keep to and zero-retention promises. Insights remains switched off.
Separately, Notion says a claim that began on 11 November that a large Notion dataset was for sale on an is false: the sample did not match its systems and had fields that real Notion data does not have.

Why it matters: A no-retention clause is only as good as the checks behind it. If your company relies on vendor ZDR, ask which features fall outside it, and if Notion emailed your , review the affected chats and change any pasted passwords or keys.

Sources (3)

Notice about the Fireworks AI incident that may impact some Notion workspaces · Notion Help Center
「AI処理、データ非保持が守られず」――Notionが委託先の問題指摘、漏えいの恐れ巡り · ITmedia AI+
False claim of Notion data for sale (November 2026) · Notion security advisories

Next · Notion securityJapanese firms say Notion AI chat leak included client and staff names →

✕Fri 13 Nov ‹›

Notion security · Fri 13 Nov

Japanese firms say Notion AI chat leak included client and staff names

99people whose details TimeTree says were in exposed chats

Several Japanese companies now say the chat leak through outside vendor Fireworks AI may have exposed client names, and staff details, ITmedia AI+ reported on 12 November. It follows Notion's own notice of 9 November (US time), which we covered on 11 November.

Key points

Calendar app maker TimeTree said on 12 November that exposed chats included personal details of 99 people: 13 survey and interview participants, 5 client contacts and 81 executives and staff, plus some business information. It has reported the case to Japan's and says no passwords were included.
Security firms SecureNavi and LRM and agriculture firm Fager published notices on 11 November after hearing from Notion on 10 November. SecureNavi says client company names, contact names and the content of enquiries and meetings may be included; LRM says the affected chats were from 5 to 24 October.
Notion's Japanese advisory says that on 11 November it Notion found inside affected users' messages. That shows some people did paste secrets into AI chats; keys and passwords for other services are not revoked automatically.
All of these companies say their own systems were not broken into, and warn customers about scam emails or calls pretending to be them or Notion.

Why it matters: Notion says fewer than 1% of were affected, but these notices show what leaked chat transcripts mean in practice: client names and meeting notes. If your workspace got Notion's email, check the chats and change any keys or passwords that were pasted into Notion AI.

Sources (4)

Next · How teams use NotionCandidly's Notion agents update product docs whenever its code changes →

✕Thu 12 Nov ‹›

How teams use Notion · Thu 12 Nov

Candidly's Notion agents update product docs whenever its code changes

1.5 hrsa week per product manager on docs (Candidly's figure)

In a short video posted on Notion's YouTube channel on 11 November, a product leader at Candidly explains how the company uses to keep its product in step with its . This is a Notion-made customer story, and the figures are Candidly's own.

Key points

To start, filled in a 'product 101' template for each feature, guided by what the speaker calls 'codebase treasure maps' so the coding agent doesn't start from zero each time. Two people then reviewed and edited each document separately, to set a human-checked baseline.
Now, every time code is merged into the main development branch, agents read the change and suggest edits to the matching documents as comments. People review only the , and their replies are logged so the agent can learn what it got right and wrong.
Candidly says each product manager now spends about an hour and a half a week on documentation. Sales staff ask Notion questions, such as which options are offered, and get answers that cite the docs and the code.

Why it matters: If your documentation always lags behind your product, the part to copy is the human-checked starting point plus reviewing only the changes. Start with one feature and keep a person approving every .

Sources (1)

How Candidly uses agents to update product docs when code changes · Notion

Next · SecurityHack at AI vendor Fireworks exposed some Notion AI chats →

✕Wed 11 Nov ‹›

Security · Wed 11 Nov

Hack at AI vendor Fireworks exposed some Notion AI chats

<1%of Notion workspaces affected (Notion)

Notion says a security incident at Fireworks AI, an outside company it used for some AI processing, exposed chat transcripts from fewer than 1% of . Notion published its notice on 9 November (US time) and says it has contacted affected customers. If you haven't heard from Notion, it says your workspace has not been identified as affected and no action is needed for now.

Key points

Notion's timeline: Fireworks first warned it on the evening of 2 November, when the stolen data turned out to be test data only. By the morning of 3 November Notion had cut its systems off from Fireworks and deleted its data there. On 6 November Fireworks said customer data may also have been taken, and it confirmed that second dataset on 8 November.
The exposed chats were copies used for , a Notion feature that shows teams how they use AI. Notion says that for most affected workspaces they were 'unattributable': not labelled with the person or workspace, and without the results of tools the AI used. But anything a person typed into the chat is still in the text.
Notion has switched off Insights while it investigates and has stopped using Fireworks to run in Notion. It says no other Notion features were affected and that this was not a breach of Notion's own systems.
Fireworks says the attackers are linked to a financially motivated criminal group behind ''. It found account and billing details were taken and, for some customers, secrets, datasets and unfinished models. It has hired Palo Alto Networks' Unit 42 to investigate.

Why it matters: If your workspace received Notion's email, review the affected chats inside Notion as instructed, treat them as exposed, and (passwords, ) that anyone pasted into Notion AI. Everyone else: treat any AI chat box like email and keep secrets out of it.

Sources (3)

Fireworks AI incident (security advisory) · Notion
Update on October 2026 security incident · Fireworks AI
AI company Fireworks says hackers stole data, billing info during October breaches · The Record

Today

Friday 13 Nov

Thursday 12 Nov

Wednesday 11 Nov

Notion says Fireworks kept AI chats despite promising zero data retention

Key points

Japanese firms say Notion AI chat leak included client and staff names

Key points

Candidly's Notion agents update product docs whenever its code changes

Key points

Hack at AI vendor Fireworks exposed some Notion AI chats

Key points