MAG212 CYBER ORIGINAL ARTICLE

AI Developer Platform Breaches: WorkOS, Trigger.dev, Chainstack and the Keys You Need to Rotate

November 5, 2026 · El Mostafa Ouchen

AI Developer Platform Breaches: Rotate the Keys You Stored

In one week, several AI developer platforms disclosed intrusions: WorkOS, Trigger.dev, Chainstack, Northflank and CommandCode. Attackers are interested in the secrets customers store with these services, especially AI-provider API keys, and Wiz says stolen keys were abused within half an hour in two cases. But what each company lost is different. Read each notice before you decide what to rotate.

Key facts

ItemDetail
ChainstackAttacked Oct 26; customer account database accessed, including password hashes; customers told to rotate node credentials (per SecurityWeek)
WorkOSFound Oct 30; internal sales/support app; CRM and support data affected; no evidence application data was accessed (per WorkOS)
Trigger.devNov 1; environment variables exposed for a subset of organizations; exploit blocked (per Trigger.dev)
Northflank, CommandCodeListed by SecurityWeek; details not confirmed in our review
WizSame keys used by several actors; first misuse within 30 minutes in two cases
Diagram of how an AI developer platform breach exposes customer secrets, with what WorkOS, Trigger.dev and Chainstack say was exposed
How a platform breach becomes the customer’s incident. MAG212 CYBER original diagram, based on vendor disclosures and SecurityWeek’s reporting of Wiz research.

What each company says

WorkOS identified unauthorized access on October 30 to an internal application used by its sales and customer-facing teams. It says the affected data is CRM and support information: business contacts, account histories, notes from sales calls, support conversations and customer Slack threads, and product usage. It says there is no evidence that WorkOS application data or production services were affected, and that the attackers used AI-based offensive tooling. Its advice to customers is to watch for phishing that uses those real details.

Trigger.dev says an exploit gave an attacker access to environment variables for a subset of organizations on November 1, that the exploit has been blocked, and that affected account owners were emailed mitigation steps. It does not say which variables were exposed.

Chainstack, attacked on October 26, had a customer account database accessed, including organization names, account holder names, emails and password hashes, according to SecurityWeek. Node credentials may also have been exposed, and customers were told to rotate them.

Why this matters even if you weren’t breached

Wiz found that several threat actors used the same stolen keys, which suggests they are being sold, and in two cases the first unauthorized activity came within 30 minutes of the credentials being exposed. Wiz links the surge to LLM-jacking: running AI workloads on stolen tokens or reselling access. SecurityWeek also reports that multiple companies link the open-source Artex AI pentest framework to some of these attacks and to Storm-0940, a Chinese threat group. We could not confirm whether this is the same “ARTEX” tooling India’s CERT-In names in its Crafty Falcon advisory.

What to do

  1. Inventory. List every secret you stored with each affected vendor: AI provider keys, cloud keys, database URLs, node or RPC credentials, webhook secrets.
  2. Rotate in priority order, starting with AI provider keys and cloud credentials, and revoke the old values rather than just adding new ones.
  3. Contain cost and blast radius: one key per project, spend caps, and alerts on usage spikes, new models or new regions.
  4. Follow each vendor’s emailed instructions and status page.
  5. Warn staff that phishing may now quote real support cases and sales conversations, as WorkOS cautions.

Rotation worksheet

SecretWhere to rotateDone when
AI provider API keysProvider consoleOld key revoked; usage normal for 48 hours
Cloud access keys in env varsIAMOld key IDs absent from audit logs
Database URLs and passwordsDatabase and app configOld credentials fail; app healthy
Node / RPC credentialsProvider dashboardNew credentials deployed; old revoked
Webhook signing secretsEach integrationOld signatures rejected

For decision-makers

Ask each platform vendor where your secrets are stored, who inside the vendor can read them, and how quickly they will tell you about a breach. Prefer short-lived, workload-scoped credentials over long-lived keys in environment variables, and treat any third party that holds your keys as part of your attack surface.

Frequently asked questions

I use WorkOS. Do I need to rotate keys?

Based on WorkOS’s statement, the affected data was CRM and support information, not application data or secrets. The main risk is targeted phishing. Follow any direct instructions WorkOS sends you.

How fast do stolen AI keys get abused?

In two cases Wiz saw the first misuse within 30 minutes of exposure. Rotate first, then investigate.

Are these breaches connected?

Some are reportedly linked to the Artex framework and Storm-0940, but a single actor behind all of them has not been established.


Sources

Related coverage on MAG212 CYBER