Your Notion AI Chats Leaked: The Hidden Risk of AI Supply Chains

The Invisible Link in Your AI Workflow
When you use an AI feature inside a tool like Notion, you are rarely interacting with a single company. Behind the interface lies a complex web of model providers, cloud hosts, and infrastructure platforms. The recent breach at Fireworks AI, reported by TechCrunch, proves that your data is only as secure as the weakest link in that chain.
Fireworks AI, which provides infrastructure for developers to host and run open-source models, suffered a breach that allowed attackers to access customer billing information and datasets. Because Notion utilized Fireworks for its AI "Insights" feature, the fallout spilled over directly to Notion users.
The Scope of the Exposure
Notion confirmed that approximately 1% of its user base was affected. The stolen data includes transcripts from the "Insights" feature, capturing both the user's input and the model's output. While Notion claims much of this data was unattributable, the incident serves as a stark reminder: AI chat logs are sensitive documents that are being passed through multiple servers before they reach your screen.
This is not an isolated incident for Fireworks AI, which has now faced two security events in two months. According to the company, the attackers appear to be financially motivated, specifically targeting access to LLMs to sell unauthorized tokens for model usage.
How to Protect Your Workflow
On November 3, Notion severed its connection with Fireworks AI and purged the data stored on those systems. For the average professional, this highlights the necessity of treating AI chat interfaces as public-facing storage.
- Audit your inputs: Avoid pasting sensitive internal strategy, unreleased financial data, or personally identifiable information (PII) into AI features that rely on third-party infrastructure.
- Check for updates: When a major service like Notion experiences a breach, look for official disclosures regarding exactly which features were compromised.
- Assume zero-trust: Even if you trust the primary software provider, assume that any data sent to an AI agent could be processed or stored by a third-party infrastructure partner.
Verdict: Is your data safe?
The convenience of integrated AI often masks the underlying complexity of data routing. If your role involves handling confidential client data or proprietary code, the risk of a supply-chain leak is real. Until AI providers offer more transparency regarding their infrastructure partners, the safest approach is to compartmentalize your most sensitive work away from AI-enabled features.
Sources
Frequently asked questions
- Was all Notion data leaked?
- No. Notion stated that the breach affected approximately 1% of its customers, specifically those who used the 'Insights' feature.
- What kind of data was exposed?
- The attackers accessed chat transcripts from the Notion AI 'Insights' feature, which includes the content of both user prompts and model responses.
AI Briefing in 3 Minutes
Get what actually matters in AI & tech delivered free to your inbox. Zero spam, 1-click unsubscribe.
Comments
0 commentsDeixe seu comentário
Be the first to comment.


