Safety

This course is about using AI without getting hurt by it. It covers what you may share and how far to trust an answer, and what can go wrong once an agent is allowed to act on your behalf.

0%

1. getting started
2. halfway
3. almost there
4. complete

explanationWhat may go into an AI tooltutorialWhat an invented fact looks liketutorialSeeing bias across many answersexplanationWhen the tool is usually righttutorialAsking a model to disagreetutorialSizing an agent's blast radiustutorialReading agent logs and rerunning after a model change

tutorialRedacting a document before you paste ittutorialSaying that AI helped, and crediting what it copiedtutorialChecking habits that run when you are in a hurry

tutorialFollowing a claim back to its sourcetutorialChecking what an agent changed

explanationMarking what a person has checkedexplanationWhy agent safety is different

tutorialTracing a planted instruction to the tool that leakstutorialWriting an AI policy a team can applytutorialAssessing the risk of a use case before it startsexplanationIntroduction to the EU AI Act

Review: nothing due yet

About this course

Goals

• Uses AI safely with confidential and personal data
• Judges the risk of letting an agent act
• Verifies AI output before relying on it

Counts

18 lessons, 53 checkpoints

Builds on

• What an agent is
• Capabilities and limits

Using AI safely, and judging the risk of letting an agent act. Progress is kept in this browser; see your progress.

Lesson plan (18 lessons, 18 live)

#	Lesson	Status	Mode	Minutes	Covers	Serves	Exercise	After	Issue
Responsible use
1	What may go into an AI toolsafety/responsible-use	live	explanation	35 estimate	responsible-use	decides-what-to-sharediscloses-ai-userespects-licenses	do
2	Redacting a document before you paste itsafety/redact-before-you-paste	live	tutorial	30 estimate	responsible-use	decides-what-to-share	do	What may go into an AI tool
3	Saying that AI helped, and crediting what it copiedsafety/saying-ai-helped	live	tutorial	25 estimate	responsible-use	discloses-ai-userespects-licenses	do	What may go into an AI tool
Recognizing failure
4	What an invented fact looks likesafety/spotting-hallucination	live	tutorial	25 estimate	failure-modes	checks-claims	judge
5	Seeing bias across many answerssafety/bias-in-patterns	live	tutorial	30 estimate	failure-modes	calibrates-trust	judge	What an invented fact looks like
6	When the tool is usually rightsafety/when-the-tool-is-usually-right	live	explanation	20 estimate	failure-modes	calibrates-trust	judge	What an invented fact looks like
Verifying outputs
7	Checking habits that run when you are in a hurrysafety/checking-habits	live	tutorial	25 estimate	verification	checks-claimscalibrates-trust	do	What an invented fact looks like, When the tool is usually right	#120
8	Following a claim back to its sourcesafety/following-the-source	live	tutorial	20 estimate	verification	checks-claims	judge	Checking habits that run when you are in a hurry	#122
9	Asking a model to disagreesafety/asking-a-model-to-disagree	live	tutorial	20 estimate	verification	spots-sycophancy	do	Checking habits that run when you are in a hurry	#124
10	Marking what a person has checkedsafety/marking-what-a-person-checked	live	explanation	35 estimate	verification	checks-claimskeeps-a-check-habit	judge	Following a claim back to its source	#126
11	Checking what an agent changedsafety/checking-what-an-agent-changed	live	tutorial	20 estimate	verification	calibrates-trust	do	Checking habits that run when you are in a hurry	#128
Agent risk
12	Why agent safety is differentsafety/agent-risk	live	explanation	30 estimate	agent-risk	names-blast-radiuschooses-human-in-looprecognizes-injection	do
13	Sizing an agent's blast radiussafety/sizing-the-blast-radius	live	tutorial	20 estimate	agent-risk	names-blast-radiuschooses-human-in-loop	do	Why agent safety is different, Checking what an agent changed	#130
14	Tracing a planted instruction to the tool that leakssafety/tracing-a-planted-instruction	live	tutorial	20 estimate	agent-risk	recognizes-injection	judge	Why agent safety is different	#132
Governance and oversight
15	Writing an AI policy a team can applysafety/writing-an-ai-policy	live	tutorial	25 estimate	governance	sets-oversight	do	Why agent safety is different	#135
16	Reading agent logs and rerunning after a model changesafety/reading-agent-logs	live	tutorial	25 estimate	governance	sets-oversight	do	Writing an AI policy a team can apply	#138
17	Assessing the risk of a use case before it startssafety/assessing-a-use-case	live	tutorial	25 estimate	governance	sets-oversightnames-blast-radius	do	Writing an AI policy a team can apply	#141
18	Introduction to the EU AI Actsafety/eu-ai-act	live	explanation	30 estimate	governance	sets-oversightdiscloses-ai-use	judge	Why agent safety is different