AI security news · 3 min read

A breach at an AI vendor exposed customers’ chats.

Notion says AI chat transcripts from its Insights feature were exposed after attackers reached AI provider Fireworks AI through a third-party cloud provider.

A breach at an AI vendor exposed customers’ chats.: HanulSec guide
Key points
  • Fireworks AI was accessed through a third-party cloud provider on 23 October 2026.
  • Notion says under 1% of users were affected and most transcripts could not be tied to a person.
  • Notion cut traffic to Fireworks, had the vendor’s copy deleted and disabled the dataset.
  • Notion advised users to rotate any passwords or API keys typed into AI chats.
A breach at an AI vendor exposed customers’ chats.: HanulSec AI Security Brief
Fireworks AI · Notion · the one-page brief we shared on LinkedIn.

What happened

BleepingComputer reported on 11 November 2026 that Notion notified users about a breach at Fireworks AI, a provider it used for AI features. Attackers reached Fireworks through a third-party cloud provider on 23 October. Notion says AI chat transcripts from its Insights feature were exposed, that fewer than 1% of users were affected and that most transcripts could not be linked to a person. Notion stopped sending traffic to Fireworks, had the vendor delete its copy and disabled the dataset.

Who is affected

  • Companies whose AI features send prompts, chats or logs to outside model hosts
  • Teams using analytics, evaluation or fine-tuning vendors for AI data
  • Staff who paste passwords, keys or personal data into AI chats

Why it matters

AI features quietly add new data processors. Prompts, chat logs and fine-tuning data often flow to model hosts and tools that never went through a full security review, and those vendors have vendors of their own. Your AI risk includes all of them.

What to do now

  • List every vendor that receives prompts, outputs or logs, and what each one keeps.
  • Shorten retention and mask personal data before it leaves your systems.
  • Tell staff that secrets do not belong in prompts, and rotate any that were shared.
  • Add AI vendors to your incident response plan and contracts.

How HanulSec helps

  • We map where your prompts, chats and logs travel.
  • We test what each AI vendor can see and keep.
  • We help you cut what is not needed and plan the response before a vendor calls you.
Want us to test yours?

Tell us what you run and we will send a fixed-price proposal within one business day. Request a quote.

Questions

Quick answers.

Should we stop using AI vendors?
No, but treat them like any vendor that handles sensitive data: know what they keep, for how long, and who else they share it with.
What should employees do?
Avoid pasting passwords, API keys or personal data into AI chats, and rotate anything that was shared.
Request a quote

Scope it in two minutes.

Tell us what you need protected. A security engineer, not a salesperson, replies with a fixed-price proposal.

  • Reply within one business day
  • Free 30-minute scoping call
  • Mutual NDA before any detail is shared
  • Fixed price, free retest of fixed findings