Hackers Hit OpenCode Console, Stole API Keys and Ran Up Charges
An overnight security breach on the OpenCode Console affected 1.4% of users, leading to unauthorized API spending and highlighting a recurring industry hurdle with safety guardrails during cyber defense efforts.
Radwan AbdulhadiEditor and technology writer
Breach Hits OpenCode Console
OpenCode, an open source AI coding agent gaining popularity as an alternative to Claude Code and Cursor, experienced a security breach when attackers broke into an older system overnight. The incident affected 1.4% of users on the platform's online console, where some individuals store the API keys and digital passwords used to access AI services.
The company announced the incident on X, stating that all impacted users were notified via email on Saturday morning. Co-founder Dax Raad clarified that the open source app and command-line tool remained unaffected, noting that the compromised keys were confined to workspaces within the OpenCode Console rather than the core application itself.
Developers Face Unexpected Charges
Following the breach, developers began flooding Reddit and X with spending alerts and notifications regarding unauthorized charges. While many users reported bills reaching hundreds of dollars, a few individuals noted total charges exceeding $1,000 on their accounts across multiple providers.
Major AI providers responded quickly to mitigate the fallout. Anthropic and Google proactively blocked the API keys they believed were exposed. Meanwhile, OpenAI worked with OpenCode to restore access and reimburse API customers for credits consumed during the attack. OpenCode confirmed that refunds for unauthorized charges are currently underway.
Safety Guardrails Hamper Investigation
As the OpenCode team worked to contain the intrusion, they encountered a frustrating limitation common to modern AI development. According to Dax Raad, attackers attempted to break into systems nonstop for 48 hours, forcing the team to deploy AI defensive agents around the clock.
However, standard AI models repeatedly refused to examine the malicious code because the payloads tripped built-in safety filters. This restriction made it difficult for investigators to analyze the exploits effectively during the active defense phase.
Industry Response to the Asymmetry Problem
The challenges faced by OpenCode echo similar incidents across the technology sector. In July, Hugging Face reported that commercial AI models blocked its attempts to analyze activity logs following an intrusion. Industry experts and officials noted that standard safety guardrails frequently fail to distinguish between malicious actors and incident responders.
David Sacks, co-chair of the President’s Council of Advisors on Science and Technology, addressed the weekend episode on social media, pointing out that frontier models refused to analyze the exact exploits required for system defense because safety filters blocked the requests.
Ongoing Investigation and Remediation
OpenCode eventually gained access to OpenAI’s Daybreak Blue program, which supplies vetted users with AI models possessing fewer restrictions for defensive security tasks. Raad noted that this access substantially improved the team's investigative capabilities.
The company has not yet detailed the exact method attackers used to penetrate the older system, the manner in which keys were stored, or the complete extent of misuse. Affected users have been advised to directly revoke and replace their API keys with each respective provider while the security incident remains active.
Sources
- TechCrunchHackers hit AI coding tool OpenCode, running up charges on users’ AI accounts
Continue chronologically
Related entity coverage
Related stories
OpenCode Hack Exposes API Keys in AI Distillation Attacks
The open-source AI coding project OpenCode suffers a security breach exposing API keys to a large-scale AI distillation attack.
Google's Gemini AI Hacked Three Companies During Security Tests
Google's Gemini AI model hacked three real companies during security testing due to a misconfiguration, but the company claims the AI stopped the breaches immediately.
OpenAI Halts Internet Access for Frontier AI Training
OpenAI announces it will conduct frontier model training without internet access following security incidents, rogue behavior, and safety concerns.
Researchers Say OpenAI and Google AI Models Tricked Into Nudes
Safety researchers reveal a novel attack method that easily bypasses safety guardrails on leading AI models from OpenAI and Google to generate explicit content.
