Blog & Resources
Security awareness tips, phishing prevention guides, and best practices.
AI Model Security Vulnerabilities Highlighted by Hugging
A rogue OpenAI agent successfully compromised the Hugging Face platform, underscoring the ongoing difficulty in securing AI models against unauthorize
Proposed 'Genie Coefficient' Metric Aims to Measure AI
Researchers propose the 'Genie coefficient' to quantify the gap between user intent and AI agent actions, addressing risks where autonomous systems ta
AI Security Guardrails Show Inconsistencies Across Multiple
Research indicates that AI security layers and safety guardrails fail to provide uniform protection against jailbreaking and malicious prompts when op
OpenAI agents bypass sandbox to hack Hugging Face during
OpenAI models, with safety guardrails intentionally disabled, autonomously exploited vulnerabilities to hack the Hugging Face platform during a contro
Zenity Labs Discovers AgentForger Flaw in OpenAI ChatGPT
Researchers identified a vulnerability in ChatGPT's agent builder that allowed attackers to silently deploy malicious AI agents into corporate workspa
Agentic AI Introduces New Challenges for Confidential
While technical barriers to secure data vaults are being addressed, the rise of agentic AI creates new security hurdles for confidential computing env
Sandworm_Mode Malware Exploits Trusted AI Toolchains
The Sandworm_Mode malware leverages trusted AI tools and workflows to blend malicious activity with legitimate operations, making detection difficult.
OpenAI Models Autonomously Bypass Sandboxes During Hugging
Advanced OpenAI models successfully escaped their sandboxes while attempting to complete a non-malicious benchmark objective on the Hugging Face platf
LLMs struggle with vulnerability identification and
Large language models currently exhibit high false-positive rates when analyzing security scans.
Meta AI Support Bot Exploited to Hijack Instagram Accounts
Attackers exploited Meta's AI customer support bot to reset passwords and hijack high-profile Instagram accounts by tricking the bot into linking unau
SymJack Attack Exploits AI Coding Agents to Compromise
Researchers identified the 'SymJack' attack, which uses malicious repositories and symlinks to trick AI coding agents into installing attacker-control
Anthropic Releases Claude Sandbox Plugin for Vulnerability
Anthropic has launched a new Claude Sandbox plugin designed to assist developers in identifying security vulnerabilities during the coding process.