Hackers Exploit OpenCode Console Vulnerability, Racking Up Unauthorized AI Service Charges

Image: TechCrunch · Source

OpenCode Console suffered a security breach compromising API keys, leading to unauthorized AI usage charges for 1.4% of users. The incident highlights challenges defending against AI-driven cyberattacks due to current model safety restrictions.

OpenCode, an open-source AI coding agent competing with tools like Claude Code and Cursor, experienced a security breach affecting approximately 1.4% of users on its online platform, OpenCode Console. This console stores API keys that users employ to access various AI services. Attackers infiltrated an older system overnight, gaining access to some users’ API credentials.

The breach resulted in unauthorized usage of AI models tied to the stolen keys, with many users reporting substantial unexpected charges ranging from hundreds to over a thousand dollars. These charges appeared across major AI providers including OpenAI, Anthropic, and Google. Prompt responses from these companies involved blocking exposed API keys and collaborating with OpenCode to restore normal operations and refund victims.

OpenCode clarified that the breach was isolated to the Console platform and did not affect the open-source app or command-line tools, which do not hold users' keys. The company notified affected users via email and advised key revocation and replacement with individual AI service providers.

During the ongoing attack, OpenCode’s team faced relentless intrusion attempts, necessitating continuous deployment of AI-based defensive agents. However, they encountered obstacles as several AI models initially used for defense refused to analyze malicious code snippets due to embedded safety filters, complicating incident response efforts. Access to OpenAI’s Daybreak Blue program—offering vetted AI models with relaxed restrictions—later aided in investigation and mitigation.

This event underscores a larger issue impacting security researchers and companies using AI: safety guardrails designed to block harmful content can inadvertently hinder forensic analysis and defense during cyberattacks. Similar limitations have been noted previously by firms like Hugging Face. Experts argue that current AI model restrictions can create asymmetries where attackers exploit fewer constraints while defenders face barriers.

OpenCode has yet to disclose specific details about the attack vector or the full scope of the breach. The company and AI providers continue working to contain the incident. This breach emerges amid a rise in AI-enabled cyber threats globally, including recent attacks leveraging AI penetration testing tools against financial institutions.

Overall, the OpenCode Console breach illustrates the growing cybersecurity challenges at the intersection of AI adoption and evolving threat landscapes, emphasizing the need for improved defense frameworks that balance operational safety and incident investigation capabilities.