The risk map: Cheatsheet
The core idea
Section titled “The core idea”The risks of generative AI sort into four groups. Once they are sorted, each becomes something you can name and watch for rather than a single cloud of dread. Against the four squares sit the levers for managing them: rules, technical tools, and the oldest lever of all, education.
If you remember one thing: The risks of AI fit on one page, and every square has a lever. Knowing the levers is the difference between fearing the technology and steering it. A map is a tool, not a warning.
The four squares
Section titled “The four squares”| Square | Everyday examples | Your move |
|---|---|---|
| What it gets wrong | False or fabricated answers, bias, leaked private data | Treat a fluent answer as a draft, not a verdict; verify anything that would cost you if it were wrong |
| What people make it do | Scams, impersonation, fraud (the voice-cloning distress call) | Recognition plus resilience: call back on a known number, agree on a family code word |
| What it changes slowly | Shifts in jobs, in relationships, in the skills we keep | Notice the trade as you make it; keep the skills you actually want to keep |
| The tail risk | Rare, large-scale, long-horizon harm | Know it is a live area of serious full-time work; read the safety track; do not let it swallow the map |
The levers
Section titled “The levers”| Lever | In plain words |
|---|---|
| Disclosure | Telling the public when content is AI-made |
| Registration | Telling the government what systems are in use |
| Licensing | Requiring approval before a system ships |
| Auditing | Checking that systems meet the rules |
| Watermarking | A hidden mark embedded in AI content so it can be traced |
| Detection | Software that tries to spot AI-made content after the fact |
| Education | Making people sharper readers of what they see |
No lever is a silver bullet. Watermarks help only when responsible companies add them; detection is a race between the fakers and the catchers; disclosure works on the honest and does nothing to the dishonest. Anyone who tells you a single fix will catch every AI-made fake is selling something.
The three-question policy test
Section titled “The three-question policy test”| Question | What it checks |
|---|---|
| Would it achieve the goal? | Effectiveness, and at what unintended cost |
| Can we build and run it in practice? | Technical and logistical feasibility |
| Is there enough support to enact it? | Whether it can actually pass |
A good policy has to pass all three, not just sound good.
The voice-cloning distress scam
Section titled “The voice-cloning distress scam”| Recognition | Resilience |
|---|---|
| A call using a short voice clip lifted from social media sounds like a relative in tears, saying there has been an accident and they need money now | Slow down when urgency and fear rush you toward money or data |
| It works because the fake sounds real and fear does the rest | Hang up and call the person back on a number you had before the call |
| Counted in United States fraud reporting for 2025 among AI-boosted scams that together cost victims roughly 900 million dollars, older adults hit hardest | Agree on a family code word a real relative would know and a cloned voice would not |
Safety is a choice, not a property
Section titled “Safety is a choice, not a property”A model refuses because people ran a tuning step that chose which requests to turn down. That is a values choice, so three questions are fair to carry:
- Who does the tuning?
- Who audits it?
- Whose values get encoded when the two disagree?
Lesson 1 flagged this as one of the hardest open problems in the field. Knowing it is a values question, and not a purely technical one, is the fluency.
The rules being written now (freshness, hedged)
Section titled “The rules being written now (freshness, hedged)”When the course was taught in 2024, formal rules for AI were mostly proposals. As of mid-2026 some are law. The European Union’s AI Act sorts systems by how risky their use is and, from August 2026, requires that people be told when they are talking to a machine and that certain AI-generated content be labeled. Its heavier obligations for high-risk uses, such as AI in hiring or credit decisions, were pushed back during 2026, with the main deadlines now landing in late 2027 and 2028. The durable point is not any single date. It is that the levers are no longer hypothetical. These dates are current status, not permanent law, and get re-checked on a rolling basis.
Pitfalls
Section titled “Pitfalls”| Pitfall | Correction |
|---|---|
| Collapsing the four risks into one dread | The everyday false-answer risk and the long-horizon tail risk are not the same size or shape; keep the squares separate and spend your attention where the everyday risks live |
| Trusting a lever too much | Watermarking, disclosure, and detection all help and none is airtight; no single fix will catch every AI-made fake |
| Reading a refusal as proof of safety | A refusal shows tuning, not a guarantee; guardrails are speed bumps a determined person can sometimes get around, which is why alignment is called unsolved |
One-liners
Section titled “One-liners”| Line | Meaning |
|---|---|
| A map is a tool, not a warning | You do not have to be afraid of a place you can find on a chart |
| Four squares, one map | The four risk categories on a single page |
| A refusal is a speed bump, not a wall | Safety tuning is real but not airtight, which is why alignment is unsolved |
| Safety is a choice, not a property | A model refuses because people tuned it to, using values someone picked |
| Every square has a lever | Knowing the levers is the difference between fearing the technology and steering it |