Notion is a widely used app for notes, documents and shared team workspaces, with its own AI assistant, Notion AI. Its own systems were not breached. Some customers’ AI chats were exposed by a company Notion used to run AI models, which many of them will never have heard of. Any organisation using AI features in everyday tools has suppliers like this.
What happened
Fireworks AI runs AI models for other companies on its own platform. Notion used it to run some open-weight AI models and to process chats for Insights. Insights is a Notion feature that shows customers how their teams use Notion AI.
Fireworks has disclosed two incidents. In the first, found on 6 October 2026, someone used internal Fireworks credentials. They captured API keys and tokens that some customers had stored with Fireworks. In the second, a third-party cloud provider hosting Fireworks’ research machines had a security incident on 23 October. The next day, the attacker used that access to reach Fireworks’ development machines.
Fireworks’ update of 9 November says billing information, resource details and some customers’ datasets were accessed. It links the activity to a group that sells stolen access to AI models.
Notion’s advisory sets out what happened next:
- 2 November: Fireworks told Notion a small dataset had been accessed. Notion found it held only synthetic test data.
- 3 November: As a precaution, Notion disconnected from Fireworks and deleted all its data there.
- 6 November: Fireworks said customer data may also have been accessed.
- 8 November: Fireworks confirmed the final dataset. It included Notion AI chat transcripts used for Insights.
- 9 November: Notion contacted affected customers.
Notion says fewer than 1% of workspaces are affected. Most transcripts were not labelled with the writer or workspace. The Register points out that the chats themselves could still contain identifying details. Notion has turned off Insights and no longer runs models through Fireworks. It asks affected customers to treat the data as exposed. Anyone who pasted passwords, API keys or other secrets into Notion AI during the incident should change them.
The Register, citing Bloomberg, reports that Fireworks is also working with Cursor and Harvey on the impact.
What to have in place
The lesson is about your suppliers’ suppliers. These RiskWitness Blueprint requirements are the ones this story tests.
- RW-SUP-13 Subcontractors known and held to the same standard (Advanced). Critical suppliers should tell you which subcontractors they use for your services. For AI features, that includes who runs the models.
- RW-SUP-04 Agreements in place before personal data is shared (Foundation). A data processing agreement normally lists approved subcontractors. Check whether your AI tools’ agreements do.
- RW-SUP-08 Supplier changes assessed (Established). Adding an AI feature can bring in a new subcontractor. Treat that as a change to assess.
- RW-SUP-16 Only approved online services for work (Established). Know which AI services staff use for work information, and approve them first.
- RW-GOV-03 Agreed rules for using systems and information (Foundation). Tell staff plainly what they may put into AI tools. Passwords and keys should never go in.
- RW-SUP-15 Critical suppliers in your incident plans (Advanced). Agree how a supplier will tell you about an incident at its own suppliers.
How RiskWitness helps
“Which of our tools send data to an AI provider?” Suppliers and Assets. Record each tool as a supplier with its criticality. Link it to the assets it touches. Send a questionnaire asking which AI providers process your data, what they keep and for how long. Store the answers and evidence, with a review date.
“Are our contracts good enough?” Suppliers. Record the contract clauses for each tool, such as breach notice times and approved subcontractors. Gaps become actions.
“What is the risk, and who owns it?” Risks. Add a risk that a supplier’s AI provider exposes your data, with an owner. Score it and choose a treatment.
“What will we change?” Controls, Policies and Actions. Add controls such as an approved list of AI services. Add staff rules on what goes into them. Link your acceptable use policy. Map each control to the requirements above, and record tests and their results. Give each gap an action with an owner and a due date.
“Do we need to tell anyone?” Incidents. If a supplier notifies you, record the incident and who is handling it, and link the supplier. RiskWitness shows which regulators may need telling, with their deadlines.
“What do we tell leadership?” Reports. Produce a PDF report or slide deck.
Every step is kept in a history that shows who did what and when, with the old and new values. Nobody can edit or delete it. When leadership, an auditor, a customer or a regulator asks how you manage your suppliers’ suppliers, the answer is already written down.
Sources
- Fireworks AI incident, Notion, updated 11 November 2026
- October 2026 security incident: update, Fireworks AI, 9 November 2026
- October 2026 security incident, Fireworks AI, updated 10 October 2026
- Fireworks AI gets burned twice as customer data exposure widens, The Register, 10 November 2026
- Fireworks AI discloses new breach after cloud provider hack, BleepingComputer, 10 November 2026
