TOPIC 04 OF 05

Prompt injection and unsafe tool use

Content can contain instructions
PLAIN-LANGUAGE IDEA

A connected AI may read webpages, emails, documents, or tool results that contain instructions attempting to override its task. If the system can access data or take actions, untrusted content can become an indirect route to those capabilities.

SEE IT IN A SITUATION
A research agent opens a webpage containing hidden text telling it to ignore the user, reveal stored information, and send a message. The page is data to analyze, not an authority that may redefine the task.
DO NOT MISS THIS

A stronger system prompt is not a complete defense. Separate data from instructions and restrict tools, destinations, credentials, and high-impact actions outside the model.

TRY THIS NOW

Constrain permissions, isolate untrusted content, validate tool inputs and outputs, require approval for consequential actions, and test adversarial examples.

CONNECTED LANDMARKS
Important topics change.

Environmental impact, copyright, privacy, security, and EU obligations require dated sources and context. The Atlas teaches durable decision habits and clearly separates them from legal or professional advice.