Northflank Breach Caps Black October Run on AI Developer Keys

Key Takeaways

• Northflank disclosed a breach on October 30, 2026, and told customers to assume every stored secret was decrypted.
• Customers reported unauthorized use of their stored AI API keys the same day.
• Wiz counted at least nine AI developer platform incidents in October and called the run Black October.
• In two cases Wiz traced, stolen keys drove about 2.5 million Bedrock calls worth over $100,000.

Stored AI Keys Become Someone Else's Compute Bill

Cloud platform Northflank disclosed a breach after an alert from Wiz. BleepingComputer reported that customers soon saw their stored AI API keys used without permission, the latest in a month of similar attacks.

What We Know

Wiz alerted Northflank at 11:43 UTC on October 30. Northflank patched within hours and told customers to rotate every key, then confirmed reports of AI key misuse.

In a November 2 update summarized by the Cloud Security Alliance, Northflank said the attack targeted control plane secrets and was contained within about an hour. No root cause has been published.

Wiz lists nine platforms with October incidents, including Fireworks AI, Railway, OpenCode, Warp, and Northflank. The OpenCode and Warp incidents are already on the PointGuard AI Security Incident Tracker.

What Happened

The intrusion reached a control plane that could decrypt customer secrets. What followed made it an AI incident.

Wiz traced two stolen Amazon Bedrock keys into use. One made about 800,000 model calls over 30 hours; the other about 1.7 million over 13 hours.

Both keys were tested within 30 minutes of exposure, and one appeared in a reseller pool a week later. This pattern, called LLMjacking, works because a model key converts directly into costly compute.

Why It Matters

The cost is real: Wiz estimates over $100,000 in Bedrock usage across two cases, billed to the key owners.

Developers increasingly park AI keys on hosting and coding platforms. Nine of those platforms reported incidents in one month, and Northflank customers lost keys through no fault of their own.

Vendor secrets handling is now part of AI supply chain due diligence. The NIST AI RMF expects organizations to manage that third-party risk.

PointGuard AI Perspective

Black October shows that a long-lived key that can call a model is one of the most valuable assets in an AI stack. PointGuard AI helps shrink that exposure.

The PointGuard AI Gateway gives applications and agents a governed path to model providers, where usage is attributable and policy applies, instead of raw keys in every environment.

PointGuard AI brokers agent secrets at runtime through HashiCorp Vault, AWS Secrets Manager, or Azure Key Vault, using short-lived, scoped credentials. A stolen environment variable becomes far less useful.

With Agent Identity and Access Security, every agent authenticates with its own identity and short-lived OAuth 2.0 tokens, so credential use is traceable and revocable.

Trustworthy AI adoption means treating AI credentials as high-value identities: brokered, scoped, monitored, and quickly revoked.

Incident Scorecard Details

Total AISSI Score: 7.80/10

Criticality: 8, Stored secrets included AI provider and cloud credentials. AISSI weighting: 25%

Propagation: 7, The same targeting hit nine platforms and spread to resellers. AISSI weighting: 20%

Exploitability: 8, Confirmed decryption of secrets and high-volume key abuse. AISSI weighting: 15%

Supply Chain: 8, Keys were exposed through third-party platforms outside customers' control. AISSI weighting: 15%

Business Impact: 8, Confirmed unauthorized usage and more than $100,000 in estimated charges. AISSI weighting: 25%

Sources

Third-Party Sources

• BleepingComputer: Northflank security breach leads to abuse of customers' AI API keys
• Cloud Security Alliance: Black October Reaches PaaS

PointGuard AI Sources

• PointGuard AI Security Incident Tracker
• PointGuard AI Gateway
• PointGuard AI Agent Identity and Access Security