Black October: Nine AI Developer Platforms Breached as Hackers Hunt for AI Keys

October 2026 turned into one of the worst months on record for the platforms developers use to build and run AI applications. Cloud security company Wiz counted at least nine AI developer platforms that disclosed security incidents during the month, and in its report “Black October” for AI Developer Platforms it describes one common thread: attackers going after the stored environment variables where customers keep their secrets, especially the keys that unlock paid AI models.

The latest name on the list is Northflank, a cloud hosting platform for apps, containers, databases and AI workloads. On Oct. 30 it told customers to assume every environment variable on its platform had been decrypted and to rotate all of their keys, and it later confirmed that customers were reporting unauthorized use of their AI API keys, BleepingComputer reported. For any business that plugs AI into its products, the message is simple: the keys you hand to a third-party platform are now a prime target, and stolen keys turn into real bills fast.

Nine Platforms in Less Than a Month

Wiz’s list, by date of incident, runs from Fireworks AI on Oct. 6 through Andromeda, Zeabur, Lightning AI, OpenCode, Railway, Warp and CommandCode to Northflank on Oct. 30. Wiz says Fireworks, OpenCode and Warp all reported signs of an “agent swarm” attack, meaning automated AI agents doing much of the intrusion work, while Andromeda called its incident part of an “industry-wide attack.” Wiz has not said that one group is behind all nine, and it found no single vulnerability or software dependency linking them.

The root causes reported so far vary. According to Wiz, Fireworks, Zeabur and Lightning AI traced their incidents to credential access, Railway pointed to privilege escalation in its build infrastructure, Warp suggested a third-party vendor dependency, and OpenCode blamed a legacy system. The consequences have been serious. Wiz says Zeabur is shutting down because of its incident, and Lightning AI revoked all user access two hours after spotting unauthorized credential access on Oct. 15, according to the incident tracker pwn.watch.

What Happened at Northflank and Railway

Northflank’s own incident page says it received a threat alert from Wiz on Oct. 30 and pushed a network policy patch soon after, followed by more fixes. It has since enabled credential rotation for PostgreSQL, MongoDB, MySQL and Redis add-ons and turned on audit logs for all customers, including free accounts. It told users to revoke credentials with the issuing providers, check AI provider usage and billing logs for unexpected activity, and report suspicious charges. BleepingComputer noted that Northflank has not said how attackers got in or how many customers were affected.

Railway, a hosting platform the security firm MATT says has about 5 million users, published a preliminary report saying an attacker accessed credentials for its CI environment and ran an unauthorized query of encrypted variables, “of which a subset were decrypted.” Railway put the exposure at “at most 1.85% of all workspaces,” which MATT estimates at about 100,000 workspaces. Railway paused new sign-ups, asked providers including GitHub, GitLab, Anthropic, OpenAI, OpenRouter and Stripe to reset exposed keys, and said it cannot refund customers for spending run up on compromised API keys.

Why Attackers Want AI Keys

The prize is access to expensive AI models. Wiz ties the campaign to LLMjacking, where criminals use someone else’s credentials to run AI models and leave the victim with the bill, or resell that access to others. In two cases Wiz investigated, compromised enterprise credentials were used to make about 2.5 million Amazon Bedrock API calls, generating roughly $100,000 in AI usage, according to BleepingComputer. MATT reports that in one case the stolen keys were first used within 30 minutes of the theft, and that OpenCode users reported overnight charges of up to $1,700.

There is a second motive too. OpenCode and CommandCode both attributed their incidents to attempted “distillation,” Wiz says, which means using a stronger model’s answers to train a cheaper rival model. Wiz also found infrastructure linked to services that resell access to AI models, though it could not tie those operations directly to the breaches. A ready market for stolen keys makes this kind of theft easy to cash in, which helps explain why so many platforms were hit in such a short span.

Why It Matters for Business Readers

Your AI keys are a financial liability. An AI API key is effectively a corporate credit card with no spending limit unless you set one. If a hosting platform you rely on is breached, the charges land on your account, and as Railway’s statement shows, the platform may not cover them. Set hard spending caps and usage alerts with every AI provider you use.

Third-party risk now includes developer tools. Most of these incidents did not start with the victims’ own systems. They started at the platforms that store customer secrets. Security and procurement teams should ask hosting, CI and AI tooling vendors how secrets are encrypted, who can decrypt them, and how fast customers are notified.

Rotation has to be fast and routine. Northflank’s advice to assume every variable was decrypted is the right default after any platform breach. Companies that keep an inventory of where each key lives, and that use short-lived or narrowly scoped credentials, can rotate in minutes instead of days. Those that do not may still be paying for an attacker’s usage weeks later.

Automated attacks raise the pace. Reports of “agent swarm” activity suggest attackers are using AI to find and exploit weak points faster than before. Defenders will need monitoring that flags unusual API usage within minutes, not at the end of a billing cycle.

The Bottom Line

Black October shows that the AI boom has created a new kind of valuable loot: the keys that pay for AI. Nine platforms disclosed breaches in under a month, at least one is closing, and customers are already facing surprise bills. Any business building on AI should treat its API keys like cash, cap what they can spend, know where every one is stored, and be ready to rotate them the moment a vendor reports trouble.

Edited by Coron @QUE.COM
Website: https://QUE.COM Intelligence
Sponsored by: https://MAJ.COM AI Autonomous