OpenCode AI Coding Tool Hit by Hackers, Users Face Charges and AI Safety Filters Block Defense Efforts

OpenCode, an open-source AI coding agent popular as an alternative to Claude Code and Cursor, reported a security breach affecting 1.4% of users of its online platform, OpenCode Console. Hackers broke into an older system overnight Saturday, stealing API keys used to access AI services, resulting in unauthorized charges on user accounts, with some reporting totals above $1,000.

In response, Anthropic and Google blocked the exposed API keys, while OpenAI collaborated with OpenCode to restore access and refund affected API customers. OpenCode has confirmed that refunds for unauthorized charges are underway. Co-founder Dax Raad clarified that the exposed keys were confined to Console workspaces, assuring that the core open-source app and command-line tool were not affected.

The incident highlighted a growing challenge in cybersecurity: the difficulty for companies without vetted access to AI models with fewer safety restrictions to defend against large-scale, AI-driven cyberattacks. The OpenCode team faced continuous attempts to breach their systems for 48 hours. They encountered significant frustration when AI models employed for defensive analysis refused to examine malicious code, triggering safety filters. This issue, dubbed "The asymmetry problem" by Hugging Face after a similar incident in July, underscores how guardrails can impede incident responders.

David Sacks, co-chair of the President’s Council of Advisors on Science and Technology, commented on the situation, stating that "frontier models refused to analyze the very exploits needed for defense because the payloads tripped safety filters. The guardrails backfired." OpenCode later gained access to OpenAI’s Daybreak Blue program, which provides vetted users with AI models less prone to refusing defensive security requests, significantly improving their investigative capabilities. The breach occurred on the same night OpenCode made its GLM-5.3-Flash model, known for its cybersecurity capabilities, free for users.

This incident adds to a recent series of AI-enabled cyber events. Cybersecurity firm CrowdStrike reported a China-based hacker likely used an AI penetration-testing agent and Claude in attacks on South Korean banks. Last week, India’s cybersecurity agency warned of AI agents reaching Indian organizations without authorization. OpenCode has not yet detailed how attackers gained access or the full extent of misuse, advising affected users to revoke and replace their keys directly with providers, noting the incident remains ongoing.

What to watch: Further details on the breach and evolving strategies for AI-powered cyber defense and model access.

Editor's note: Comprehensive coverage of the OpenCode cyber incident, incorporating the technical context, safety filter hurdles, and leadership quotes.