Topic guide · Updated
Agentic AI Security
AI agents read data, call APIs and take actions on their own. That makes them useful, and it makes them a new kind of privileged user that most security programs weren't designed for. This guide organizes my writing on the subject into five questions, each with a practical checklist you can use today.
Governance: knowing which agents exist and what they may do
Most agent risk starts before any attack: agents get deployed outside review, with broad permissions, and nobody owns them afterwards. Governance means an inventory, a risk tier for every agent, and a culture where teams report the tools they use instead of hiding them.
AI Security · 8 min
Shadow AI: When Employees Deploy Agents Faster Than Security Can Catch Them
Shadow AI is AI tools and agents deployed without security review. Learn the risks, from credential theft to prompt injection, and how to govern it.
AI Governance · 8 min
Agentic AI Oversight: A Framework for CISO Risk Classification
A three-dimension framework for classifying AI agents by data sensitivity, autonomy, and operational scope, with matching controls for each risk tier.
Security Culture · 7 min
Security Behavior Culture Programs for GenAI
Why annual awareness training fails for GenAI risks, and how to build a security behavior and culture program that reduces shadow AI and unsafe AI use.
Identity and access: least privilege for non-human actors
An agent is a machine identity that can take actions. Treat it like one: give it its own workload identity, short-lived credentials, narrowly scoped API access, and continuous verification rather than a shared service account with standing permissions.
Zero Trust · 7 min
Zero Trust for AI Workloads
How to apply zero trust to AI agents, models and data pipelines: workload identities, least privilege, data segmentation, monitoring and human approval.
Identity & Access Management · 7 min
Identity Security for Machine Actors at Scale
Machine identities outnumber human ones and get far less oversight. How to discover, rotate, scope, and own service accounts, API keys, and workload identities.
API Security · 7 min
API Security in the Age of AI Agents
AI agents call APIs at machine speed and can follow injected instructions. Secure APIs with least privilege, response sanitization, and behavior monitoring.
Detection and response: when agents or attackers misbehave
Agents change both sides of the fight. Attackers use AI to generate malware variants and convincing deepfakes; defenders use it to triage alerts. Detection has to focus on behavior rather than signatures, and response playbooks need steps for AI-specific incidents.
Security Operations · 7 min
AI-Driven SOC: Why Analysts Need New Skills in 2026
AI SOC tools only help when analysts can judge their output. The skills SOC teams need in 2026, how to build the training program, and what to measure.
Malware Defense · 7 min
AI-Driven Malware: Why Behavior-Based Detection Is the Only Way Forward
Malware like PROMPTFLUX now queries LLMs for new evasion code at runtime. Why behavior-based EDR, attack-path analysis, and credential hygiene matter more.
Emerging Threats · 7 min
Deepfake Defense: Protecting Your Organization in 2026
Deepfake video and voice fraud now targets finance, HR, and help desks. Learn the defense layers: out-of-band verification, provenance, detection, and training.
Platform and supply chain: what your agents are built on
Agents inherit every weakness of the code and devices beneath them: the dependencies they load, the firmware on edge devices they control, and the cryptography protecting their traffic. Securing the platform means knowing your components, signing what you ship, and planning for post-quantum cryptography.
Supply Chain · 8 min
Software Supply Chain Attacks Quadrupled. Here's Why.
Large supply chain compromises nearly quadrupled since 2020, per IBM X-Force. The real attack vectors, why they work, and three practices that cut your risk.
Cryptography · 7 min
Post-Quantum Cryptography Is Closer Than You Think
Post-quantum cryptography is urgent now. What NIST's FIPS 203, 204 and 205 standards cover, why harvest-now-decrypt-later matters, and how to start migrating.
Edge Security · 7 min
Edge Security for IoT and 5G Networks
How to secure IoT devices, edge servers and 5G links: unique device identity, encrypted communication, segmentation, signed firmware, and retrofitting fleets.
Exposure and resilience: finding gaps and recovering fast
Some weaknesses will always slip through, so the goal is to find exposures continuously and recover quickly when something breaks. That covers cloud misconfigurations, prioritizing what attackers can actually reach, and backups an attacker can’t encrypt or delete.
Cloud Security · 7 min
Cloud Security Posture Management: What CSPM Does and How to Evaluate It
What cloud security posture management does, how it differs from CWPP and manual audits, and how to evaluate a CSPM platform against your own cloud estate.
Vulnerability Management · 7 min
Continuous Exposure Management: The End of Quarterly Vulnerability Scans
Quarterly scans leave gaps attackers exploit. Continuous Exposure Management maps attack paths, covers identities, and prioritizes by real exploitability.
Ransomware Defense · 7 min
Ransomware Resilience: Immutable Backups and Incident Response Orchestration
How to build ransomware resilience with immutable, isolated backups and a rehearsed incident response playbook, drawing on CISA, FBI, KELA and Verizon data.
Where to start
If you're responsible for agents already in production, start with an inventory using theshadow AI audit checklist, then classify each agent with the risk classification worksheet. Everything else is easier once you know what exists and how much damage each agent could do.