A breach at Fireworks AI exposed Notion AI chat transcripts in a small number of workspaces. Here's why those conversations matter more than you think, and how to protect yours.
Notion just told its users something that should make all of us pause. A security breach at Fireworks AI, the third-party provider powering some of Notion's AI features, exposed AI chat transcripts from a small number of workspaces. No passwords, no billing details, just conversations people had with an AI assistant inside their private workspace.
Sounds minor, right? Maybe not as minor as it first appears.
### Why Chat Transcripts Are a Bigger Deal Than They Look
Think about what you actually type into an AI assistant while you work. You paste in a draft contract. You ask it to summarize a customer complaint. You brainstorm a pricing strategy you haven't told anyone about yet. Those transcripts aren't just idle chit-chat. They're a running diary of your thinking, your business, and sometimes your clients' secrets.
That's the part most people miss. We treat AI chats like a private scratchpad, but the moment a third-party model provider is involved, that scratchpad lives somewhere else too. And when that somewhere else gets breached, your words go with it.
Notion was quick to clarify that the exposure was limited. A small number of workspaces, they said, and the data didn't include account credentials. Fair enough. But "small number" is cold comfort if yours happens to be one of them.
### The Third-Party Problem Nobody Wants to Talk About
Here's the uncomfortable truth: most of the AI tools you use every day aren't running on the company's own servers. They're plugged into outside providers, and each one is another door into your data.
- Your workspace platform connects to an AI provider.
- That provider connects to infrastructure and model hosts.
- Each layer adds another place where something can go wrong.
When a breach happens two or three steps down the chain, you usually hear about it late, and you hear about it vaguely. Notion handled disclosure reasonably well, but the core lesson stands. You can't fully control what happens downstream. You can only control how much you hand over in the first place.
> The safest data is the data you never paste into a chat box.
### What You Can Do Starting Today
You don't need to panic, but you do need a few habits. First, treat every AI chat like it might be read by someone else someday. If you wouldn't email it to a stranger, don't type it into a prompt. Second, check which AI features are turned on in your workspace and whether you actually need them. Third, keep sensitive client details in separate, controlled documents rather than in casual AI conversations.
For teams that handle a lot of private data, this event is a nudge to review vendor agreements and ask the boring but important question: where exactly does our data go, and who can see it?
### The Bigger Takeaway
AI is moving into everything, and that's mostly good. But convenience always has a price, and the price here is a little more exposure than we're used to. Notion's Fireworks incident wasn't a catastrophe. It was a warning shot. The companies and individuals who take it seriously now will be the ones who avoid the messier version of this story later.
So the next time you open a chat window and start typing, ask yourself one quick thing: would I be okay if this showed up somewhere I didn't expect? If the answer is no, maybe keep it out of the prompt.