Plan this weekPlatform changes

Fireworks AI says the October attack reached billing data and some customers' models and datasets: if you use it, rotate your keys and read its notice

Fireworks AI updated its security bulletin: the attackers, who it links to an LLMjacking group, first reached its stored secrets, then billing information and, for a limited set of accounts, secrets, datasets and incomplete models, through a third-party cloud provider. What the company says and what to do.

How urgent

Plan this week

Who is affected

Fireworks AI customers, above all those notified by the company. It says it is notifying the accounts whose data was affected. Inference, Training SDK and Nexus, which keep no data by default, show no impact so far. We do not know how many accounts were affected

What to do

Read the notification email from Fireworks AI, replace every key and secret you stored on the platform and the third-party credentials behind them, check your usage and billing for activity you do not recognise, and review the datasets and models you uploaded

Published 10 Nov 2026 · Sources last checked 10 Nov 2026No. 72 / 87

What happened

Fireworks AI, a platform that runs AI models for developers, has updated the security bulletin it first published on 9 October 2026. According to the company, at that time only secretsSecret (API key, token, password) Any piece of text that proves to a service that you are allowed in, such as an API key, a database password or an access token. Tap again to open the glossary that Fireworks itself stored, such as those used in evaluation jobs, were known to have been accessed. Since then it found impact on account and billing information and, for a limited set of accounts, on secrets, datasets and incomplete models.

The company says a third-party cloud provider that hosted its research and development machines had a security incident on 23 October. On 24 October an unauthorized party used that to reach Fireworks development machines and run an unauthorized workload, which led to more customer data being affected. Fireworks says it detected this the same day and that, as of 25 October, it has not detected further successful unauthorized access to the platform customers use. It does not name the cloud provider.

Fireworks says the evidence indicates links to a financially motivated criminal group that sells access to large language models through illicit gateways, a practice known as LLMjacking, and points to a 2 November post by Google Threat Intelligence Group and Mandiant on the subject. It has called in Palo Alto Networks Unit 42 for forensic analysis and notified law enforcement.

What data was involved

According to the bulletin, the affected internal application gave access to:

  • Resource metadata: identifiers, display names, descriptions and creation times.
  • Billing information: billing and shipping addresses, invoice and payment history, account balance and credit usage. Payment card numbers are not stored by Fireworks and were not affected.
  • Model weights and datasets: for certain customers, a subset of incomplete model weights and datasets.

Fireworks says it found no impact to Inference, Training SDK or Nexus, which keep no data at rest by default (zero data retention), and that its findings are based on the investigation so far, which is ongoing. As a precaution it disabled self-serve dataset uploads and downloads for Batch Inference and dataset-based training.

Who is affected

  • Fireworks AI customers. The company says it is notifying the customers whose data was affected and giving them the details and recommended steps. If you have an account and received no email, the contact it gives is [email protected].
  • Customers who stored secrets on the platform, for example keys used in evaluation jobs: those were the first thing reached.
  • Not shown: other AI platforms. We did not look at them.

What to do

Our advice, not the company's list: it has not published specific steps beyond the ones in its customer notification.

  1. Read the notification from Fireworks AI and follow it. If you did not receive one, write to the contact above before assuming you are not affected.
  2. Replace your Fireworks keys and, more importantly, the third-party credentials you ever stored there (cloud, GitHub, model-provider keys). Deleting a stored secret does not make the credential itself stop working: the provider that issued it has to cancel it.
  3. Check usage and billing for spikes or calls you did not make. LLMjacking shows up as someone else's usage on your bill.
  4. Review the datasets and models you uploaded. If some held customer data or secrets, treat them as exposed and tell the people involved as your own obligations require.
  5. Watch for phishingPhishing A message or page that pretends to be someone you trust to get you to hand over a password, a code or a payment. Tap again to open the glossary that uses your billing details. Names and addresses can make a scam email look real.
  6. For the future: keep secrets out of evaluation jobs and datasets, give each service its own key, and cap the spend on every key.

What we could not confirm

We read Fireworks AI's updated bulletin directly, and a press report of it through a subagent. We did not read the original October bulletin, the notification emails or the Google and Mandiant post. We do not know the name of the cloud provider, how many accounts were affected, whether any secrets were actually used, or how the attackers first got in. The link to the LLMjacking group is Fireworks' reading of the evidence, not a named attribution.

Watch the short video

The video loads from YouTube (privacy-enhanced mode) only when you press play.

Comments

No account needed. Add a nickname if you like, or leave it empty to comment anonymously. Comments are reviewed before they appear. Links are not allowed.

    We store your comment, your nickname and a scrambled form of your connection that we use only to limit spam. We do not store your IP address or set cookies. Cloudflare Turnstile checks that you are not a bot.

    Sources and verification

    We use primary sources first and say when we rely on a news report. Dates are the day we last checked each source.

    Spotted a mistake? Email [email protected] and we will correct it and note the change.

    Get the free runbooks

    One-page security checklists for founders who build with AI. You get a confirmation email first, then the runbooks.

    Send me the runbooks