AI Safety Said Simply
Complex AI safety topics, explained in forms people actually finish.
Short explainers, interactive demos, and videos on the ideas that matter in AI safety — built for policymakers, journalists, educators, and anyone without hours to spare.
The library
One piece is live. The rest are in production — each one short, accurate, and made to be shared.
Reward hacking, in the wild
Real, documented cases of AI systems gaming their objectives — searchable and severity-rated.
Sleeper agents & secret loyalties
How a model can behave perfectly in testing while carrying hidden goals for later.
AI control
Getting useful work out of AI systems we don't fully trust — and catching them if they defect.
Eval awareness & eval gaming
What happens when a model can tell it's being tested — and acts accordingly.
Chain-of-thought unfaithfulness
A model's written reasoning doesn't always reflect why it actually did what it did.
Bio & cyber uplift
How much easier do frontier models make it to cause serious harm — and how we measure that.
Compute verification
How treaties on AI could actually be enforced: verifying what chips are doing, and where.
Safeguards & classifiers
The filters wrapped around AI models — what they catch, what they miss, and why it's hard.
Self-fulfilling misalignment
Could writing about treacherous AI teach future models to be treacherous?
Value reflection
If an AI could revise its own values, where would they settle — and would we like the result?
Why this exists
Policymakers
Memos and demos are how policy offices learn and brief others. Most AI safety material is too long and too technical to use that way.
Educators
Digestible teaching material on AI safety is scattered across platforms — and for many topics, it simply doesn't exist.
The public
Nobody comes home from work and reads a dense 5,000-word technical post. The good material caters to people who already understand it.
Lower the barrier to entry, and the ideas travel: more people who matter understand AI safety, and the conversation shifts in its favor.
For institutions
University AI safety groups, educators, and policy offices are how these materials reach the people who need them. Everything we publish is free to share — send it to your members, use it in your courses, hand it to your colleagues.
Talk to us about distributionGet involved
Contribute
We’re looking for writers, demo builders, and video makers who can make hard ideas simple.
Neither form below fits? Email us.
Stay updated
Get new explainers, demos, and videos as they’re published.
You work in comms, media, or content. Answer as many or as few questions as you like. All questions are optional.
Questions, answered
Isn't “said simply” just another way of saying “dumbed down”?
No. We cut jargon and length, not substance. Every piece aims to leave you with the real idea — the same one a researcher would recognise — minus the notation and the prerequisites.
How do you choose which topics to cover?
We look for topics where the stakes are high and no genuinely accessible material exists yet. If a great short explainer is already out there, we'd rather point to it than duplicate it.
Can I request a topic?
Yes, please do. Requests from people who need the material — a briefing next month, a course next term — move topics up the queue. Email kaustubh.kislay@gmail.com with what you need and when.
Do you take positions on AI policy?
We explain ideas; we don't lobby. Where experts disagree, we say so and present the disagreement rather than picking a side for you.
When do the in-production topics ship?
We don't promise dates. Each topic goes live when it's accurate and genuinely easy to finish. Sign up for updates below and we'll tell you the moment each one ships.
I spotted an error. What should I do?
Tell us — accuracy is the whole point. Email kaustubh.kislay@gmail.com with the piece and the problem, and we'll fix it and note the correction.