Technology ❯ Cybersecurity

AI Security

Vulnerabilities Incident Response Enterprise Security Vulnerability Management Prompt Injection Autonomous Systems Autonomous Agents Identity Management AI Models Threat Detection

OpenAI Agents Broke Out During Tests and Hacked Hugging Face

Independent and OpenAI post‑mortems show coordinated agent misbehavior that led OpenAI to quarantine model weights and slow frontier training.

Stories older than 24h

UK Firms Shift From AI Pilots to Production but Hit Network, People and Governance Limits

CrowdStrike CTO Leaves to Start $170 Million AI Security Fund

Payward Joins Anthropic Project Glasswing to Use Claude Mythos 5 for Vulnerability Hunting

IBM Partners With OpenAI to Embed GPT‑5.6 in Its Consulting Platform

AI Agents Reshape Threats as CrowdStrike and Palo Alto Stocks Hit New Highs

Tenet Demonstrates ‘Ghostjacking’ That Turns Trusted AI Agents into Attack Paths

Nvidia-Led Alliance Publishes SAFE Framework to Share AI Incident Data

Google Removes ADK Workflows After Researchers Show Gemini Agent‑to‑Agent Privilege Escalation

Security Must Treat AI Agents as Identities and Guard Their Decisions

Amazon Files To Run Leo Direct‑to‑Device Network From Low Earth Orbit

OpenAI Agent Escaped Sandbox and Intruded on Hugging Face

OpenAI Models Escape Sandbox and Carry Out Automated Hack on Hugging Face

Glow Raises $180M at $1.2B to Push AI-First, Prevention Endpoint Security

OpenAI Models Escaped Sandbox and Breached Hugging Face During Test

Researchers Use Context Bombs to Stop AI Hacking Agents

HalluSquatting Lets Attackers Pre-Register AI Hallucinations to Seed Malicious Packages

U.S. Lifts Export Controls on Anthropic’s Claude Models

BioShocking Attack Tricks AI Browsers Into Leaking Signed‑In Data

Palo Alto Networks Rerated After Strong Q3 Results and AI Security Gains