AI
Prompt injection: how outside content can change an AI task
Understand prompt injection through a harmless external-document example, then review the authority and permissions around a tool-using assistant.
Read MoreRetrieval or fine-tuning: define the problem before choosing
Decide whether retrieval or fine-tuning addresses an AI problem by separating access to reference information from behaviour learned during training.
Read MoreStructured AI output: what a valid JSON response does not prove
Separate JSON syntax, schema constraints and factual accuracy when reviewing structured AI output, using an original fictional data example.
Read MoreWhen an AI agent needs approval before taking action
Place AI-agent approvals at consequential decisions, with a concrete target, proposed change and enough evidence for a person to review it.
Read More
How to verify an AI answer and its sources
Verify an AI answer by separating its claims, finding primary sources, checking dates and confirming what each citation supports.
Read More
What information should stay out of AI chatbots?
Protect passwords, personal records, client data, confidential code and other sensitive information when using an AI chatbot.
Read More