InAI for Backend EngineersbyNavya Devadiga·2d agoAgent Memory Poisoning Is the Prompt Injection That Outlives the Session (OWASP ASI06)An attacker plants a belief in one session; it steers a decision in a clean session later. OWASP ASI06, and the five controls in Python.A response icon1A response icon1
InAI for Backend EngineersbyNavya Devadiga·3d agoIndirect Prompt Injection Is a Data-Flow Bug: 10 Paths Into Your Agent’s ToolsA support ticket, a lead form, a crawled page. 10 injection paths from 2026’s agent breaches, each with the Python boundary that closes it.
InAI for Backend EngineersbyNavya Devadiga·3d ago11 MCP Server Mistakes Behind This Autumn’s CVEs, Each Fixed in PythonA debugger plugin on 0.0.0.0, a token nobody checked, a shell built from config. 11 MCP CVEs from autumn 2026, each with the Python fix.A response icon2A response icon2
Eng Soon Cheah·4d agoNVIDIA Garak: the nmap of the LLM worldAs Large Language Models (LLMs) become increasingly integrated into business applications, testing their security is no longer optional…
Erwin Hermanto·5d agoPrompt Injection Attacks: How We Secured Our LLM-Powered ChatbotReal attack vectors, input sanitization, and output validation techniques we learned the hard way
Ronit Kcsai·6d agoMy AI Agent Passed Its Security Checks. It Still Followed the Attack.I ran 102 experiments against an AI support agent. What I discovered changed how I think about prompt injection, security guardrails, and…
Eng. Naveen Ganeshan·Oct 27Everyone Uses LLM Dashboards. Nobody Asked Where the Data Goes.What security teams need to stop doing, and what to do instead.
Paul Deepakraj Retinraj·Oct 24The Agent Security Kernel, Part I: Threat Model, Invariants and ArchitectureBounding authority when the model is untrusted — Part I of II. Part I covers the thesis, threat model, seven invariants, architecture and a…
