Policies Don't Fail – Context Does
Policies are static, while context is dynamic. Effective control requires understanding how intent propagates across the entire system.
Research & Analysis
Intelligence on
Intelligence on
LLM Security
Threat analysis, vulnerability research, and field notes from the frontier of AI security.
(AI) Security is about Helping Everyone Else
The actual problem: enabling teams to use AI securely at the speed they operate.
The AI Security Management Gap: From Fog to Secure-by-Default Deployment
Most CISOs cannot clearly define 'AI security' or articulate how to manage it. The path from this fog to secure-by-default AI deployment starts with understanding where security management breaks down.
Defense-in-Depth for AI Agents: Why Input Security Isn't Enough
Defense-in-depth for AI agents requires securing both input and output layers.
The Lopsided Risk Dynamic of AI Agent Deployments
AI agent deployments are creating the most lopsided risk dynamic in modern security.
86% of Organizations Delayed AI Deployments Due to Security
But the ones who ship anyway are learning painful lessons. Here's what's actually breaking.
The SIEM Blind Spot: Why Traditional Security Tools Miss LLM Attacks
When Salesforce Agentforce was compromised, traditional security tools saw valid HTTPS requests. The prompt injection exfiltrating CRM data went undetected.
The Security Problem Every SaaS Company Has But Won't Talk About
AI assistants shipped fast to stay competitive. Now companies are quietly backfilling the security controls that should have been there from day one.
LangChain Serialization Vulnerabilities: When LLMs Generate Exploits
Two critical vulnerabilities show how LLM-generated content can bypass traditional security controls and execute malicious code.
The FAFO Framework: Fast Adoption, Future Accountability
Most companies adopted the same AI security strategy: ship features fast, figure out security later. Here's why that doesn't end well.
Why Traditional Security Tools Can't Catch LLM Attacks
Traditional security looks for malformed syntax. LLM attacks exploit meaning through natural language. It's a fundamentally different attack surface.