Railway breach steals AI API keys, drives $740,000 in fraudulent charges

How the breach unfolded

Railway, a US-based cloud hosting provider for developers, reported a security breach where attackers exploited a zero-day vulnerability in its build infrastructure. The flaw allowed them to partially escape filesystem isolation on virtual machines and access files containing credentials used by build servers.

Credentials and customer impact

After gaining access to a misconfigured internal task management system called Temporal, the attackers obtained temporary cloud credentials. These were later used to access a production database holding encrypted environment variables, along with the decryption keys. Railway estimates up to 100,000 customer workspaces may have had their environment variables decrypted during the 26-hour window between October 20 and October 21.

AI API abuse and financial fallout

The attackers focused on AI API keys stored as environment variables. These keys were used to generate approximately $740,000 in fraudulent charges across AI services, likely for model distillation—a process where smaller AI models are trained using data from larger, proprietary models. Railway suspects the stolen keys were sold on the black market.

Response and remediation

Railway has since patched the zero-day vulnerability, rebuilt affected infrastructure, and rotated all compromised credentials. The company is working with AI providers to refund the fraudulent charges and has committed to covering any losses that providers refuse to reimburse.

Connection to earlier AI security risks

The incident follows recent reporting on AI-driven vulnerability exploitation, where attackers used AI tools to accelerate attacks within hours of PHP vulnerability disclosures. While that case involved speed, Railway’s breach shows how AI credentials themselves can become targets, turning a technical flaw into direct financial damage.

What this means

Lazyfounder analysis — our interpretation, not reported fact.

Railway’s breach exposes a growing risk for cloud hosting platforms: AI API keys are now prime targets. Unlike traditional credentials, these keys grant immediate, automated access to costly paid services—turning a security compromise into direct financial losses. For founders, this incident is a reminder that environment variables are not just configuration details; they are high-value assets that require the same protection as database passwords or encryption keys. The $740,000 fraud bill also shows that cloud providers may have to absorb costs beyond their own infrastructure, especially when AI services are involved.

Key takeaways

• Zero-day vulnerabilities in build infrastructure can lead to large-scale credential theft, even in multi-tenant cloud environments designed for isolation.
• AI API keys stored as environment variables are attractive targets; their exploitation can result in immediate financial losses due to fraudulent usage.
• Cloud providers may face significant unreimbursed costs if attackers use stolen keys to access third-party services like AI platforms.
• The incident highlights the need for developers to treat API keys and credentials as critical assets, requiring encryption, rotation, and monitoring.

Related on Lazyfounder

• Wiz
• More technology news
• AI tools accelerate exploitation of PHP vulnerabilities within hours of disclosure