Anomaly Innovations says breach affected about 1.4% of OpenCode Console users
Anomaly Innovations, the maker of the terminal‑based coding agent OpenCode, confirmed a large‑scale distillation attack that exposed stored OpenAI, Anthropic and Google API keys for roughly 1.4% of OpenCode Console users. The company said the breach impacted only those users who had linked external providers to their workspaces and that it notified every affected account by email the same…
Key points
- Anomaly Innovations confirmed a large‑scale distillation attack affecting roughly 1.4% of OpenCode Console users.
- The attack exposed stored OpenAI, Anthropic and Google API keys for users who had linked external providers.
- Anomaly advised revoking and reissuing all provider keys and checking billing logs for unauthorized usage.
The firm could not confirm whether the exposed keys had been used maliciously, but some developers reported unexpected charges on their OpenAI and Anthropic bills after the notification. Anomaly’s documentation already warns that OpenCode does not sandbox the agent and recommends Docker or a VM for real isolation. Its advice now mirrors standard incident response: revoke and reissue all provider keys directly through the respective dashboards, check billing and usage logs back to October 17, and set spending limits to prevent runaway costs. The breach highlights the broader risk of AI agents handling sensitive credentials and adds to recent incidents that have prompted banks and cloud providers to restrict agent access.
Anomaly Innovations, behind coding agent OpenCode, says a breach hit ~1.4% of OpenCode Console users and their API keys, in a "large scale distillation attack"
startupfortune.com · 17 October 2026
Loading the full article…
This text was published by startupfortune.com and written by Walter Schulze. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- Meta may write a $100 billion-$200 billion check for Muse Wonder, says Zuckerberg · 1 src
- Nvidia says AI agent safety should rely on execution‑environment permissions, not prompts · 1 src
- Websites block AI agents from booking, shopping, and banking · 1 src
- OpenAI and Anthropic add spending limit features to their SDKs · 1 src
- OpenAI's GPT-6 Pro remains unconfirmed as AI parallel processing concept explored · 1 src
Comments
via GitHub Discussions