peekgenticBook a call

News analysis · AI security

An AI platform breach reached customer data. What it means for the AI tools your office uses

Fireworks AI, a platform companies use to run AI models, said on Nov. 9 that an October break-in reached billing records and, for some accounts, datasets and stored keys. You've probably never heard of Fireworks. That's the point.

By Miguel GutierrezCo-founder, Peekgentic · Dallas
· 4 min read
Preview card: An AI platform breach and your office.
Fireworks AI's Nov. 9 update widened an October breach to billing data and some customer datasets.Peekgentic

In brief

Fireworks AI said on Nov. 9 that attackers who came in through a third-party cloud provider reached account and billing data and, for a limited set of customers, stored secrets, datasets and incomplete models. Security firm Wiz counts it among at least nine AI developer platforms that disclosed incidents in October. For a mid-size company, the lesson is to know which AI services sit behind the tools you use, keep your keys in as few places as possible, and put a spending cap on every AI account.

  • Fireworks AI said an October break-in through a cloud provider reached billing data and, for some accounts, datasets and stored keys.
  • Wiz counts at least nine AI developer platforms that disclosed incidents in October.
  • Attackers want AI keys to resell. Google saw over $100,000 in fraudulent charges in a single day in some cases.
  • Most mid-size companies are exposed through tools they buy. Ask which AI services sit behind them.
  • Set a spending cap on every AI account and keep each key in one place.

What happened

Fireworks AI is a platform that companies use to host and fine-tune AI models. Wiz calls this kind of company an AI developer platform. On Oct. 9 it disclosed that someone had used internal Fireworks credentials to reach environment variables in a limited number of customer accounts. Those are settings where developers store things like API keys, the passwords software uses to talk to other services. At the time, Fireworks said it had found no access to customer models, datasets or prompts.

On Nov. 9, Fireworks posted an update that widened the picture. A third-party cloud provider hosting Fireworks' research machines had its own security incident on Oct. 23. The next day, the attacker used that to reach Fireworks development machines. Fireworks says that led to:

  • Account and billing information, such as addresses, invoice and payment history, and credit usage. Fireworks says it doesn't store card numbers.
  • For a limited set of accounts, stored secrets, datasets and incomplete model weights.

Fireworks says it found no impact to its inference service, which doesn't keep prompts or outputs by default. It is working with Palo Alto Networks' Unit 42 and says the evidence points to a financially motivated group that sells stolen access to AI models. It has paused self-serve dataset uploads and downloads while it investigates.

The Fireworks timeline, in its own words

Date (2026)What Fireworks reported
Oct. 6Found unauthorized use of internal credentials; began containment
Oct. 9Published first bulletin: stored secrets in some accounts exposed
Oct. 23A third-party cloud provider hosting its research machines had a security incident
Oct. 24Attacker used that to reach development machines and run an unauthorized workload
Nov. 9Update: billing data, and for some accounts datasets, secrets and incomplete models, affected

Sources: Fireworks AI security bulletins of Oct. 9 and Nov. 9, 2026.

It wasn't only Fireworks

Wiz, a cloud security company, wrote on Oct. 30 that at least nine AI developer platforms disclosed incidents in October, and it has added more names since. Wiz says the public reports share one pattern: attackers hunting for stored keys, especially keys that grant access to AI models. One of the companies, Zeabur, said it will shut down because of its incident, according to Wiz.

Wiz's running update for Nov. 10 also says Notion gave notice of downstream impact from the Fireworks incident. We haven't seen Notion's notice ourselves, so treat that as Wiz's report. But it shows the chain: a tool you know uses a service you don't, which runs on a cloud provider neither of you picked.

You can do everything right and still be exposed by your vendor's vendor.

Why criminals want AI keys

Google's threat intelligence team and Mandiant wrote on Nov. 2 about what they call LLMjacking: stealing access to AI models and reselling it. Google says it saw fraudulent charges above $100,000 in a single day in some incidents, and up to $1 million in total at some organizations before access was cut off. Wiz described two stolen keys that ran up an estimated $100,000-plus in usage at list price between them.

So the risk isn't only that someone reads your data. It's also that someone runs up a bill on your account.

What it means for a 50-300 person company

You probably don't buy from Fireworks. You may still depend on it. Most mid-size companies buy AI inside other products: a CRM, a notes tool, an AP tool, an order-entry system. Each of those may send data to one or more AI platforms behind the scenes. That's normal. The question is whether your vendor tells you who they are.

Keys are the soft spot. If anyone on your team signed up for an AI service and pasted the key into a no-code tool, a spreadsheet add-on or a hosted app, that key now lives in two places. Every extra place is one more chance it leaks. Fireworks itself told affected customers that deleting a stored key doesn't cancel it; you have to rotate it with whoever issued it.

A spending cap is the cheapest protection there is. Google's first-line advice includes budget alerts, per-model quotas and spending caps. If a key leaks, a cap turns a $100,000 surprise into a capped one.

Where the other side has a point

Fireworks disclosed, updated and named its forensic firm. Not every vendor does. It also says the parts of its service that handle live prompts keep no data by default, which limited the damage. Zero-retention settings did their job here.

Big providers aren't immune either. The attackers came in through a cloud provider. Moving everything in-house doesn't remove that risk; it just makes it yours to manage.

The details are still coming in. Fireworks says its investigation is ongoing. Wiz says most of the October incidents are still being analyzed. Some of what's been said on social media about losses hasn't been confirmed.

This doesn't mean "don't use AI." The fix is ordinary vendor management, not a ban.

What to do this week

  1. Ask your software vendors one question: "Which AI services process our data, and will you tell us within days if one of them has an incident?" A good vendor has a subprocessor list. Our questions to ask any AI vendor covers the rest.
  2. Find the AI keys your company owns. Who signed up, where the key is stored, and what it can reach. Keep each key in one place.
  3. Put a spending cap and a billing alert on every AI account. Google recommends it. It takes minutes.
  4. Check data retention. Ask whether prompts and files are kept, and for how long. Shorter is safer.
  5. Look at what your staff signed up for on their own. Our pieces on shadow AI and third-party apps in Microsoft 365 walk through how.

Common questions

What happened in the Fireworks AI breach?

Fireworks AI said on Nov. 9, 2026 that an attacker came in through a third-party cloud provider on Oct. 24 and reached account and billing information and, for a limited set of accounts, stored secrets, datasets and incomplete model weights. It reported an earlier credential incident on Oct. 9.

Were prompts or chats exposed in the Fireworks breach?

Fireworks says it has found no impact to its inference service, Training SDK or Nexus, which keep no data by default. Its investigation is ongoing.

How many AI platforms were hacked in October 2026?

Wiz counted at least nine AI developer platforms that disclosed incidents in October 2026 and has added more since. It says the reports share a focus on stealing stored keys that grant AI access.

What is LLMjacking?

Stealing access to AI models, usually through leaked keys or cloud accounts, and using or reselling it. Google says some victims saw over $100,000 in fraudulent charges in a day and up to $1 million in total.

Does this affect companies that don't use Fireworks?

It can. Fireworks is an AI developer platform, so software you buy may run on it or on a service like it. Wiz reports that Notion gave notice of downstream impact. Ask your vendors which AI services process your data.

How can a mid-size company protect itself?

Ask vendors for their AI subprocessors and incident notice terms, keep each AI key in one place, rotate keys with the issuer if a service is breached, set spending caps and billing alerts, and prefer short or zero data retention.

Sources

  1. Fireworks AI, Update on October 2026 security incident, Nov. 9, 2026
  2. Fireworks AI, Fireworks October 2026 security incident, updated Oct. 10, 2026
  3. Wiz, "Black October" for AI Developer Platforms, Oct. 30, 2026 (updated through Nov. 10)
  4. Google Threat Intelligence Group and Mandiant, The Rise of LLMJacking, Nov. 2, 2026

More from Peekgentic

Published November 10, 2026 · Written by Miguel Gutierrez, Co-Founder, Peekgentic (Dallas, TX). Every statistic links its source.