How a Rogue Prompt Could Turn Microsoft Copilot Into a $250,000 Wire Fraud Accomplice
A proof-of-concept from Barracuda researchers shows how attackers could weaponize Microsoft Copilot itself to escalate a single compromised inbox into full CEO account takeover and a quarter-million-dollar wire...
Claude Broke Out of a Sandboxed Security Test and Hit Three Real Companies, Anthropic Admits
Anthropic says a review of 141,000 evaluation transcripts turned up three cases where Claude models, told they were operating in an isolated capture-the-flag simulation, instead reached the live...
Hugging Face Breach Reveals a New Front: AI Agents Attacking, AI Agents Defending
Hugging Face has confirmed a production infrastructure intrusion driven by an autonomous AI agent, exploiting two flaws in its dataset processing pipeline. The company's own AI-based forensic analysis...
Inside NadMesh: The Shodan-Powered Botnet Hunting Exposed AI Servers
Researchers at XLab have identified NadMesh, a Go-based botnet that uses Shodan to hunt down exposed AI and MCP infrastructure before hijacking it with more than 20 exploitation...
Rogue Agent: Critical GCP Dialogflow Flaw Let Attackers Inject Malicious Code Into AI Chatbots
Varonis Threat Labs disclosed a critical Dialogflow CX flaw, dubbed Rogue Agent, that let attackers with a single edit permission inject persistent malicious code into shared chatbot execution...
New T3MP3ST Framework Turns AI Coding Agents Into Autonomous 0-Day Hunters
T3MP3ST, a new open-source framework, turns AI coding agents like Claude Code and Codex into autonomous red-teaming operators, claiming strong results on benchmark suites and a set of...
Malicious ClawHub Skills Compromise AI Agents With Hidden Backdoors — 247,000 Installs, $2.3M Stolen
Researchers scanning 50,000 ClawHub skills — the official marketplace for the OpenClaw AI agent platform — found working remote control backdoors, credential stealers, and autonomous malware that installs...
DifyTap: Critical Flaws in AI Platform Dify Allow Silent Wiretapping of AI Conversations Across 1M+ Apps
Researchers at Zafran disclosed four vulnerabilities in Dify — including two critical CVSS 9+ flaws — that let attackers silently intercept AI conversations across tenants, access private files,...