PinnedArsh Riz·Aug 5Your Copilot Agent Just Exfiltrated Payroll Data.The real AI security failure happens at ingestion, months before runtime. Here’s how to get left of bang with the Microsoft stack you…
PinnedArsh Riz·Aug 23, 2025From Principle to Pattern: Zero Trust for Agentic AI on AzureThere’s a moment every CISO dreads: Friday, 4:37 p.m., the floor is quiet, and an innocuous “helper” bot tries to open a change window on a…
PinnedArsh Riz·Dec 4, 2024Evolving Security Champions Programs in the Age of GenAI and Cloud SecurityIntroduction
PinnedArsh Riz·Oct 30, 2023Leveraging the Cloud to Securely Harness the Potential of Generative AI: A Comprehensive Guide for…The emergence of powerful generative AI models like DALL-E, GPT-3,GPT-4 and Codex provides intriguing opportunities for enterprises to…A response icon1A response icon1
Arsh Riz·2d agoAI Security Engineering: Why Multi-Turn Attacks Beat Single-Prompt FiltersPost 2 of 8 in a series on AI security engineering. Post 1 built a threat model for an email-to-CRM agent. This one looks at why the input…
Arsh Riz·2d agoAI security engineering Series — Threat Modeling an Email-to-CRM Agent: Follow the Data, Not the…Post 1 of 8 in a series on AI security engineering. One working system, its threats, its controls, and the tests that prove the controls…A response icon1A response icon1
Arsh Riz·2d agoAI Security Engineering Series: Stop Asking the Model Nicely: Enforce Output in CodePost 3 of 8 in a series on AI security engineering. Post 2 showed why input filters miss attacks. This post is about what to do after the…
Arsh Riz·2d agoAI Security Engineering Series: Interpretability for Security Teams: What’s Real, What’s ResearchPost 8 of 8 in a series on AI security engineering. The first seven posts covered controls you can deploy today. This one looks at a field…
Arsh Riz·2d agoAI Security Engineering Series: Jailbreak Tests as a Merge GatePost 5 of 8 in a series on AI security engineering. Posts 1 to 4 covered threats and controls. This one is about proving the controls still…
Arsh Riz·2d agoAI Security Engineering Series: Keeping Customer Data Out of Third-Party ModelsPost 4 of 8 in a series on AI security engineering. Posts 1 to 3 covered threats, attacks and output controls. This one is about what data…