Red Teaming a Clinical-Admin Agent Without Sending a Single Prompt Off-Site"Lakeshore Care Network" is a fictional regional healthcare group. To reduce administrative load, they built a small system of cooperating AI agents: Scheduling agent: books and reschedules appointme2h ago · 8 min readvivekchhediya.hashnode.dev
The Model Upgrade That Changed Nothing in the Code and Everything in the Behavior"Harbor Freight Logic" is a fictional mid-size logistics firm. Its AI agent helps dispatchers by: Reading shipment requests. Proposing routing and carrier assignments. Booking pickups through a carri2h ago · 7 min readvivekchhediya.hashnode.dev
One New Tool, One Big Boundary, The Refund Agent That Could Do Too Much"Cedar & Co." is a fictional online retailer. Their customer assistant started life as a friendly, read-only helper. It could: Look up order status. Explain return policies. Summarize shipping delays4h ago · 7 min readvivekchhediya.hashnode.dev
The Support Agent That Obeyed a Web PageMeet "Northwind Support," a fictional mid-size SaaS company with a lean support team and a growing ticket backlog. To cope, they built an AI agent with four capabilities: Read incoming tickets and cu4h ago · 8 min readvivekchhediya.hashnode.dev
Indirect Prompt Injection: The Attack Your AI Agent Can’t SeeIndirect prompt injection hides malicious instructions inside content an agent retrieves, such as a web page, email, document, or tool output. The agent reads it as data but may follow it as a command5h ago · 3 min readvivekchhediya.hashnode.dev
