9 AI developer platforms had breaches in October

Wiz, a cloud firm, reports that 9 or more AI developer platforms had breaches. The breaches came from 6 October to 30 October 2026. Attackers went for environment variables, which hold the secret keys for AI access. Fireworks and Railway are 2 of the 9. One platform will stop work because of its breach. Wiz says the pattern fits attacks by AI agents and a market for AI keys.

How it works

3 platforms say that attackers got credentials. 1 platform says the cause was a fault in its system. 1 says a vendor dependency, and 1 says a legacy system. Wiz saw the attackers test each key with 1 call from a home address. They then tried many models and checked the quotas of the account.

The numbers

In attack A, an actor made about 800,000 calls in 30 hours. The calls went to Bedrock, which runs AI models in the cloud. About 90% worked. In attack B, one address made about 1.7 million calls in 13 hours. Wiz says the 2 attacks likely used more than $100,000 of AI models at the usual price.

What is not known

Most breaches are in analysis. No one fault links them. Wiz cannot prove that the resellers it found caused the first breaches. Reports of $371,000 in stolen cryptocurrency come only from social media posts.

Insight: Stolen AI keys have a ready buyer

Wiz saw 2 stolen keys used by many actors, and one key appeared in a reseller pool. A ServiceNow study found that only 23% have one owner for agent rules. A market for AI keys seems to make a vendor system cash for attackers. The next move to look for is more attacks on vendors. Leaders must count the keys that vendors hold for them.