Fireworks AI hit by second attack in weeks, exposing customer data

The platform that runs AI for more than 10,000 companies warned customers of a new attack that came through a cloud provider. What it teaches about the vendor chain.

Fireworks AI, an American platform that hosts and runs AI models for companies, warned customers of a second hacker attack in just a few weeks, according to Bloomberg and Forbes. The new incident happened on October 24: the attacker exploited a flaw at an external cloud provider that Fireworks used. For a limited group of accounts, access secrets, databases and unfinished models were affected.

One attack on top of another

The company has more than 10,000 customers, is backed by Nvidia and was valued at US$ 17.5 billion in July. In early October, it had already published an official notice about the first incident: internal credentials used without authorization, with API keys and tokens stored as secrets in some customer accounts being captured. The investigation into that first case also found impact on customers’ account and billing data.

According to Forbes, two customers said they had moved much of their work to other vendors. Fireworks says it is investigating with help from Palo Alto Networks, that it has notified the authorities and that it is in direct contact with customers.

What this changes for your company

1. The risk is in the chain, not just in the vendor you hired. Your company may never have signed anything with Fireworks and still use software that runs AI on platforms like it. Ask every vendor that uses AI with your data for the list of who processes that data behind the scenes, and record it in your vendor inventory.

2. An API key is a password. The first attack started with internal credentials, and both hit customers’ access secrets. Every key that connects your systems to an AI service needs a named owner, least-privilege access and periodic rotation. And when a vendor reports an incident, rotating the key is mandatory, even with no sign of misuse.

3. A contract with a notice deadline. Customer data sent to an AI service remains your company’s responsibility. Require in the contract a deadline for the vendor to report incidents, including those that happen in its own chain, and define who decides, on your side, what to do. If you want to gauge where your company stands, the free self-assessment takes two minutes.

In 29 years of programming, I’ve seen that the hardest incident to handle is the one that happens in a system nobody at the company knew existed. With AI, that list of systems grows fast, and someone needs to own it.


Originally published on IWA7 Insights, IWA7 Digital’s blog, which is now part of Escovando Bytes.

Keep readingSee all →