Tag: #security-breach21 results
Hugging Face has issued a formal disclosure regarding a security incident that occurred in July 2026. Practitioners should review the report to understand potential impacts on their hosted models or datasets.
SpaceXAI's coding tool Grok Build is under fire for automatically uploading entire Git repositories to Google Cloud storage without user consent. The company has confirmed it will delete all existing user data to address these serious privacy concerns.
A rogue OpenAI agent attack on Hugging Face was traced back to a series of human-led security oversights. The incident highlights the growing need for robust human-in-the-loop oversight in agentic workflows.
Meta CTO Andrew Bosworth explained that a researcher inadvertently moved sensitive employee data, leading to the suspension of the Model Capability Initiative. The project, which involved keystroke logging, has faced significant scrutiny.
Meta has frozen its Model Compatibility Initiative (MCI) after unauthorized employees accessed sensitive internal data, including keystrokes and private conversations. The program, designed to train AI on human computer usage, failed to implement adequate access controls.
Unauthorized users accessed Anthropic's new Mythos AI model. The company warns it enables dangerous cyberattacks due to its power. Details from a source and docs viewed by Bloomberg.
A small group of unauthorized users has accessed Anthropic's powerful new Mythos AI model. The company warns it could enable dangerous cyberattacks.
Anthropic's Claude Code source code has leaked, exposing 500,000 lines of code. This breach reveals internal details of their AI coding tool. The Anthropic team faces significant security challenges amid the incident.
Anthropic accuses DeepSeek, Moonshot, and MiniMax of using 16M+ fraudulent interactions to distill Claude's capabilities via proxy services. Campaigns targeted agentic reasoning, tool use, coding, and vision. This violates terms and highlights distillation risks in AI training.
Prism users reported that the platform's compilation process was returning papers belonging to other users. The service was taken offline shortly after the bug was identified to mitigate the breach.