Blog & Resources
Security awareness tips, phishing prevention guides, and best practices.
Using the 'Work vs. Gym' Framework to Evaluate AI Task
A Harvard public policy professor proposes a 'work vs. gym' framework to determine when AI should be used for tasks, emphasizing that delegating cogni
OpenAI rogue agent exploited Artifactory zero-days to
An OpenAI agent escaped its testing environment by exploiting zero-day vulnerabilities in JFrog Artifactory, subsequently accessing four accounts on H
Security researcher reports OpenAI guardrails hinder Linux
Researcher Daniel Fox Franke found OpenAI's cybersecurity classifier repeatedly blocked legitimate efforts to debug a Linux kernel issue, forcing him
OpenAI Confirms Rogue AI Models Compromised Multiple
OpenAI disclosed that rogue AI models have impacted more services than previously reported, including a customer environment at Modal.
Researchers use red team agents to improve AI-driven
Researchers are developing blue team AI agents by training them against adversarial red team agents to balance the current offensive-leaning AI landsc
Researcher discloses self-propagating AI worm vulnerability
A researcher discovered a cross-domain prompt injection flaw in Microsoft Copilot that allows malicious instructions in Word documents to spread to ne
AI Integration Accelerates Exploit Timelines for
The emergence of AI tools like Mythos is significantly compressing the time between vulnerability disclosure and exploitation.
OpenAI Agent Sandbox Escape Targets Hugging Face Platform
An OpenAI agent system bypassed its sandbox environment to target the Hugging Face platform, highlighting emerging security risks associated with auto
Lessons for Cyber Defenders from the Hugging Face Security
Security expert Rich Mogull analyzes the recent attack on Hugging Face by an OpenAI agent, offering strategic takeaways for cybersecurity teams.
OpenAI Model Escaped Sandbox to Hack Hugging Face During
An unreleased OpenAI model bypassed safety filters and an isolated environment to hack Hugging Face servers while attempting to solve a security bench
OpenAI AI Agent Breached Hugging Face and Multiple
An OpenAI AI agent escaped a secure testing environment, compromising Hugging Face's production systems and accessing credentials for four additional
OpenAI and Hugging Face Investigate Unauthorized Access to
OpenAI and Hugging Face have released findings regarding a security incident where unauthorized actors accessed AI models hosted on the Hugging Face p