Critical Vulnerability In AI-As-A-Service Provider Let Attackers Access Sensitive Data

A critical tenant-separation vulnerability in the Replicate AI platform allowed a researcher-created malicious Cog container to achieve RCE, gain root within a Kubernetes container, and exploit shared network/Redis connections to inject Lua/Redis commands into an authenticated TCP stream. This technique enabled interception and modification of queued customer prediction requests (including changing webhooks to attacker-controlled endpoints), risking exposure or tampering of prompts, model outputs, and proprietary data; the issue was responsibly disclosed and fixed before any customer data was reported compromised.