The development platform Railway has confirmed that a cyberattack allowed hackers to access some customer keys, prompting the company to urge all users to rotate their API and cloud keys to prevent potential broader exposure of customer data.
The company allows customers to build applications and services on its cloud development platform. In a blog post on Wednesday, Railway stated that unidentified hackers "may have accessed encrypted environment variables and obtained decrypted contents for some of them."
Environment variables typically consist of a set of passwords and API keys that developers' applications use to access other services, such as cloud data storage and databases. With these customer keys, attackers may also log in and access data as if they were the legitimate owners.
Railway stated that although its initial analysis indicated the breach was limited to "a subset of users," the company still recommends that "all users" revoke and rotate all keys to prevent potential misuse by attackers.
The company also stated that it has suspended customer deployments during the investigation, effectively freezing customers' websites and applications.
Railway did not specify how many users were affected or the root cause of the incident. The company also did not immediately respond to requests for comment on the incident or indicate whether it had received any contact from hackers, such as ransom demands.
It is currently unclear how the attack occurred or which of Railway’s systems were compromised. Railway stated that this breach was likely part of a broader campaign targeting platforms “designed to break into AI inference providers.” Such attacks typically involve malicious prompts that trick AI systems into bypassing safeguards and leaking sensitive data.
This latest breach occurred shortly after the U.S. cybersecurity agency CISA warned that Chinese-based AI companies are launching large-scale attacks against U.S. businesses to enhance their own model capabilities.
Earlier this week, the open-source AI coding tool OpenCode stated that during a similar attack, customer keys were compromised, and hackers stole API keys and other credentials from over 1% of customers. OpenCode attributed the attack to an AI distillation attack, in which numerous agents attempt to extract sensitive information from AI models.
The attack on Railway appears unrelated to two critical security vulnerabilities in the PHP web programming language disclosed this week. Out of concern that attackers might exploit these vulnerabilities before they are patched, companies and governments worldwide have taken websites offline this week to apply fixes. (PHP is used by approximately 70% of websites and public-facing internet systems.)
The patch for the vulnerability codenamed "PHP2Shell" was publicly released on Wednesday, following global coordination to allow large organizations to patch their systems before potential attacks. Amazon security researchers said they found evidence of the vulnerability being exploited four days before the patch was released, indicating that these vulnerabilities are now being used in active attacks.