Code Is Cheap. Authority Isn't.The SDLC is changing, and so is the org chart. When the unit of engineering shifts from a developer to a human-accountable system of agents, the bottleneck moves from writing code to granting authorit2AAOct 18 · 18 min readarunima-chaudhuri.hashnode.dev
Your Agent Forgot the Prompt Injection. Its Memory Didn't.Once AI agents share memory, prompt injection stops being an input-filtering problem and becomes a chain-of-custody problem. Custody answers it with a FalkorDB graph traversal. On Monday, one of yourOct 16 · 46 min readarunima-chaudhuri.hashnode.dev
The Agent Never Opened the Email: Engineering Test Verdicts That Know What the Agent Actually SawExposure tracking, causal ordering and the inconclusive status in AgentSec 0.8–0.9, and the edge cases that turned up while building them Here's a test that looks fine. An email assistant is asked to26AKSOct 12 · 24 min readarunima-chaudhuri.hashnode.dev
What Does a Passing Agent Security Test Actually Prove?A research note on observability, harness validity and evaluator precision, using AgentSec's runs against seven agent frameworks, two memory stores and the Claude Code CLI Abstract Agent security tooOct 7 · 23 min readarunima-chaudhuri.hashnode.dev
Allowed Isn't Authorized: Testing What Your AI Agent Does With the Permissions It Already HasA hands-on guide to AgentSec 0.7's authority checks (task scope, data flow, honest reporting, identity boundaries and multi-agent delegation), with real output from reference agents, seven frameworks,36KSOct 6 · 24 min readarunima-chaudhuri.hashnode.dev
