Saturday, 31 January 2026

What can go wrong with Agents?


🔐 What Can Go Wrong with Agents 

1️⃣ Perception - Prompt injection, poisoned data, fake documents, malicious payloads
2️⃣ Reasoning - Hallucinations, bias, unsafe logic, policy violations
3️⃣ Planning - Unsafe actions, privilege escalation, illegal sequences
4️⃣ Action - Data leaks, fraud, destructive actions
5️⃣ Reflection = Poisoned memory, corrupted learning, policy drift 


What can go wrong with Agents?

🔐 What Can Go Wrong with Agents  1️⃣ Perception - Prompt injection, poisoned data, fake documents, malicious payloads 2️⃣ Reasoning - Hallu...