How do we
secure AI?
Models are one part of the system. Data, retrieval, identity, tools and orchestration all shape its risk.
AI assurance · Red teaming · Agent controlsAI Security · Autonomous Systems · Enterprise Risk
Perspectives by Vijay RaviAI is gaining agency.
Security must keep pace.
Notes on using AI in security workflows and managing the risks it introduces—from incident triage and compliance evidence to agents that can change a system.
Browse focus areasFrom the notebook
Agent control plane / Runtime security
Inside an Agent Security Control Plane: From Delegated Authority to Execution
How to preserve business intent as AI agents move from investigating a problem to changing enterprise systems.
Read the essayA system-level view
IdentityAuthority
AI AGENTReason · Plan · Act
ExecutionEvidence
Capability becomes useful when its actions remain accountable.
My focus areas
How do we
How can AI
improve security?
Moving from assistance to useful automation in investigation, remediation and cyber risk workflows.
Agentic SOC · Exposure · Risk operationsHow do we
govern autonomy?
Connecting business impact and accountability to permissions, approvals and evidence at runtime.
Ownership · Oversight · Enterprise riskA practical starting point
From a useful pilot
to a security workflow
teams can rely on.
Secure adoption requires more than a model evaluation or a governance policy. It requires a working connection between what a system can do, what it is permitted to do, and what people can verify.
Invite me to share these ideas