Making AI Headlines · Page 5
OpenCode's Unwanted Feature: Someone Else's AI Bill
· New York time
Hackers breached part of OpenCode's online platform, stole some users' API keys, and ran up charges on their AI accounts. OpenCode is an open source coding agent that has become an alternative to Claude Code and Cursor. The incident shows how access credentials can turn a development tool into a route to somebody else's paid services. The reporting also says AI tools consulted for help repeatedly refused to examine the malicious code. That creates an awkward distinction between refusing to assist wrongdoing and refusing to inspect evidence of wrongdoing. A safety boundary has to account for the purpose of an investigation, not merely the alarming appearance of the material. For users, the central issue is not which assistant sounds smartest. It is what an exposed key permits, how abuse becomes visible, and how quickly access can be contained. Those questions belong beside feature comparisons, not underneath them. The industry's dream is an assistant that handles the tedious parts of software work. The less marketable reality is that tedious security work remains essential. A tool that can explain a complicated program still needs a credential system that does not let strangers put their computing errands on the customer's tab.
Read this story in the full newspaper →Satire. Not factual reporting. Read the original reporting from the linked publishers.