AI Agent Security & Safety
Articles about AI security, agent safety, and guardrails for autonomous systems
Game-Based Prompt Injection Tricks AI Browsers Into Ignoring Safety Guardrails
LayerX researchers demonstrate how the BioShocking attack tricks agentic browsers using themed web games to bypass safety controls, successfully compromising products like Claude's Chrome plugin and ChatGPT Atlas.
The Interconnected Web of AI: How Hybrid Environments Reshape Human Agency and Society
Exploring how AI-mediated environments create cascading effects across cognition, relationships, climate, and equity—and why proactive engagement through hybrid intelligence frameworks is essential for preserving human agency.
One-Click Data Theft: How SearchLeak Turns Microsoft Copilot Into an Exfiltration Weapon
Varonis Threat Labs uncovered SearchLeak (CVE-2026-42824), a critical three-stage vulnerability chain in Microsoft 365 Copilot Enterprise that lets attackers steal mailbox, OneDrive, and SharePoint data with a single click on a trusted microsoft.com link.
AutoJack: How a Localhost Bypass Turned AutoGen Studio Into a Remote Code Execution Gateway
AutoJack is a three-flaw vulnerability chain in Microsoft AutoGen Studio enabling remote code execution via AI agents through localhost trust and unauthenticated MCP endpoints.
The ClawHub Breach: How Malicious AI Skills Are Weaponizing Agent Autonomy
Brynn Nguyen argues that the recent ClawHub malicious skill discovery highlights the severe vulnerabilities in AI marketplaces, urging a shift from static scanners to rigorous provenance verification and runtime behavior monitoring.
Langflow Path Traversal Vulnerability CVE-2026-5027 Actively Exploited in Attacks
Attackers are actively exploiting a high-severity path traversal vulnerability in Langflow, the popular AI development platform. The flaw allows unauthenticated attackers to write arbitrary files on exposed servers, potentially leading to remote code execution. This article details the vulnerability, its exploitation timeline, and necessary mitigation steps.
The Miasma Worm: A Self-Replicating Supply Chain Attack Targeting AI Coding Agents
The Miasma worm is a credential-stealing supply chain attack framework that has compromised 73 Microsoft-affiliated GitHub repositories. This autonomous worm uses compromised developer credentials to spread through CI/CD pipelines and specifically targets environments with automated AI tools.
Claw Patrol: A Security Firewall for Autonomous AI Agents
Claw Patrol provides a runtime security layer for AI agents, allowing developers to set guardrails and monitor agent behavior. Created by the Deno team, it solves the 'rogue agent' problem in autonomous workflows through protocol-level inspection and rule-based validation.
Securing Autonomous Agents: The New CISO Challenge
As agentic AI adoption accelerates in enterprises, securing these autonomous agents presents a major challenge due to their ability to act outside traditional control frameworks.
Prompt Injection and Data Exfiltration in Copilot Search
Understand the SearchLeak attack: A three-stage indirect prompt injection threat in AI-powered search. Learn detection and mitigation strategies for Secure AI deployment.