After OpenCode and Railway, Command Code has also been hacked: hackers are targeting AI model keys.

Beating AI Flash News, October 29 — AI coding tool Command Code announced it has blocked a security incident linked to a large-scale distillation attack targeting Anthropic and OpenAI models. The company preemptively revoked all its API keys, requiring users to re-login to generate new ones. Over the past two weeks, OpenCode and Railway have successively encountered similar attacks. On October 17, OpenCode’s cloud console was breached, affecting roughly 1.4% of its console users; some third-party model keys may have been leaked, with some users reporting unauthorized charges of hundreds to thousands of dollars. On October 21, Railway was also hacked, leaving up to around 100,000 workspaces at risk of environment variable leaks. Investigators found from the attackers’ code logs that hackers specifically targeted AI service API keys. Railway assessed the attackers likely sought to use these keys to extensively call models and collect outputs for model distillation. Earlier, on October 14, Zeabur was also breached. The attackers focused on searching for AI API keys, even attempting to use stolen AWS credentials to access Amazon Bedrock, but the attempt failed.