7 critical findings to 0, approved in the room: red-teaming AI agents on SwitchWhen one agent's message can make another agent issue a refund nobody asked for, where do you catch it? Inside the chat room — with a human in the loop.8 min read·arunima-chaudhuri.hashnode.devNov 6
Code Is Cheap. Authority Isn't.Agentic AI makes code cheap but authority and verification scarce. A framework for redesigning the SDLC and engineering org around autonomous agents.18 min read·arunima-chaudhuri.hashnode.dev4AAOct 18
Your Agent Forgot the Prompt Injection. Its Memory Didn't.Once AI agents share memory, prompt injection becomes a chain-of-custody problem, and a graph traversal is what proves an action is unsafe.46 min read·arunima-chaudhuri.hashnode.devOct 16
The Agent Never Opened the Email: Engineering Test Verdicts That Know What the Agent Actually SawYour AI agent test passed, but did the agent ever see the attack? How AgentSec tracks delivery, checks causality, and reports inconclusive runs.24 min read·arunima-chaudhuri.hashnode.dev26AKSOct 12
What Does a Passing Agent Security Test Actually Prove?What does a passing AI agent security test prove? Evidence from 7 frameworks, 2 memory stores and Claude Code on observability and false passes.23 min read·arunima-chaudhuri.hashnode.devOct 7