Hackers steal OpenCode Console API keys from 1.4% of users
Attackers broke into an older OpenCode system overnight, exposing API keys for 1.4% of OpenCode Console users, with some reporting more than $1,700 in unauthorized charges as of 19 October 2026.
Photo: TechCrunchKey points
Hackers breached an older OpenCode system overnight, exposing API keys for 1.4% of OpenCode Console users and driving unauthorized charges above $1,700.
Hackers broke into an older OpenCode system overnight, exposing API keys belonging to 1.4% of users of OpenCode Console, the platform where developers store digital passwords used to access AI services. OpenCode announced the breach on Saturday and said it had emailed all affected users that morning. Users who did not receive an email were not impacted, the company said.
The breach matters because OpenCode is a popular open-source AI coding agent, an alternative to Claude Code and Cursor, and because leaked API keys let attackers run up real charges on victims' accounts. Some users reported unauthorized charges in the hundreds of dollars, with a few totals above $1,000 and others reaching up to $1,700. OpenCode Console, not the open-source app or command-line tool, was affected.
What the breach exposed
OpenCode framed the incident as a large-scale distillation attack, a technique that trains smaller AI models using responses generated by more powerful ones. The technique becomes an attack when it involves large-scale training through unauthorized queries. Skeptics called the framing an overstretch, arguing the incident was a straightforward breach whose use for distillation remains unproven. OpenCode did not specify how the distillation attack occurred.
The email sent to impacted users said the earliest possible exposure of their keys was on 17 October 2026, 00:30 UTC, and that no data appears to have been exposed. OpenCode said it was working with Google, OpenAI, and Anthropic to address the incident but could not share further details at that time. Anthropic and Google blocked the API keys they believed were exposed.
Once attackers had valid credentials, developers began posting spending alerts and unauthorized charges. One developer wrote that their OpenAI, Anthropic, and Google AI spend alerts all fired off before going to bed. Some users reported Google suspended their projects after detecting abusive activity consistent with hijacked resources. OpenAI worked with OpenCode to restore access and refund API customers for credits used in the attack.
How the attack worked
Co-founder Dax Raad said attackers had been trying to break into OpenCode's systems nonstop for 48 hours using AI agents, requiring the team to keep AI defensive agents running around the clock. Raad stressed the breach concerned OpenCode Console, not OpenCode itself, writing that those keys never come to us. OpenCode has not detailed how attackers gained access to the older system or how keys were stored.
AI models that OpenCode turned to for analysis helped for a while, then started refusing to work when the malicious code they were asked to examine activated safety filters. Raad called the refusals a bit frustrating. The situation stabilised after OpenCode received access to OpenAI's Daybreak Blue program, which gives vetted users AI models less likely to refuse requests for defensive security work.
The same problem hit Hugging Face in July, when commercial AI models blocked its attempts to analyse activity logs from an intrusion by autonomous AI agents. In a blog post section titled The asymmetry problem, Hugging Face wrote that guardrails cannot distinguish an incident responder from an attacker. Security researchers have raised similar complaints, saying restrictions are inconsistent and have pushed some toward foreign open-weight models.
Why AI models blocked defenders
David Sacks, co-chair of the President's Council of Advisors on Science and Technology, responded on X, writing that OpenCode and earlier Hugging Face hit the same wall mid-attack because frontier models refused to analyse the very exploits needed for defense when payloads tripped safety filters. He wrote that the guardrails backfired.
The breach occurred the same night OpenCode made GLM-5.3-Flash free for users. Researchers who study open-weight models note the broader GLM-5.3 family has cybersecurity capabilities approaching levels previously associated with restricted-access models like Anthropic's Claude Mythos Preview. OpenCode told affected users to revoke and replace their keys directly with each provider, and said refunds for unauthorized charges are underway. The company said the incident remains ongoing. Users whose keys were exposed should expect to rotate credentials with Anthropic, OpenAI, and Google before resuming normal workloads on the Console platform.
Frequently asked questions
How many OpenCode users were affected by the breach?
Attackers broke into an older OpenCode system, affecting 1.4% of users of OpenCode Console, where some users store API keys used to access AI services. OpenCode emailed all affected users; those who did not receive an email were not impacted.
How much did unauthorized charges cost affected users?
Some users reported charges in the hundreds of dollars, with a few totals above $1,000 and some reporting more than $1,700 in unauthorized charges. OpenCode said refunds for unauthorized charges are underway, and OpenAI refunded API credits used in the attack.
Why did OpenCode's AI defensive agents stop helping during the attack?
AI models OpenCode used for analysis started refusing to work when the malicious code they were asked to examine activated safety filters. The situation stabilised after OpenCode received access to OpenAI's Daybreak Blue program for defensive security work.


