Who it is for
For people who design or scrutinize AI preparedness policy and want it to enforce real risk mitigation rather than restate corporate promises.
AI Knowledge Series / expert
The Preparedness Framework
Go beyond corporate promises: design AI preparedness policies that actually enforce risk mitigation, drawing on CSET and government frameworks.
The problem
It starts with the situation you're actually in.
Published preparedness frameworks read like binding safety commitments but function as discretionary corporate promises. You'll learn to separate a policy's stated intent from its actual enforcement power — the distinction that decides whether catastrophic frontier-AI risk is actually managed.
You've taken preparedness frameworks apart. Now you assemble one — starting from the failure mode you've seen repeatedly.
Outcomes
What you'll be able to do.
A policy is a machine made of permissions
You can read any clause in a frontier lab's safety policy as a machine of permissions. When a model crosses a dangerous capability threshold, you say exactly what the lab is required to do — not what it vaguely intends.
The idea Carnegie put a name to
A pledge to "evaluate dangerous capabilities before deployment" sounds binding until you ask the question that matters under pressure: who decides whether today's situation triggers the evaluation? If it's the same people who want to ship, you don't have a commitment.
The Ceiling Problem
You can pressure-test a governance board that looks airtight — powers separated, overrides recorded, audit log append-only — with the question that voids most of it: who pays the board's members, and what happens to revenue if they say no?
Why a maturity score, and why now
Working from the inside out, you can read a clause for what it permits, find where commitments leak, bind discretion with if-then triggers, give evaluations teeth, allocate decision rights, and plug your policy into a government backstop.
Inside the book
A closer look at the work inside.
What this book argues checklist
- Who decides that the threshold was reached? If the same leadership that wants to ship also controls whether the evaluation runs, the trigger is discretionary.
- Who decides that safeguards are "sufficient"? If "sufficient" is judged internally with no fixed criterion, the gate has a dial, not a lock.
- What happens if deployment proceeds anyway? If no one's authority is violated and no audit records the bypass, the "will not" was a "would prefer not to.".
Map the Machine Yourself
Build the reference model in your own notes. For each tracked domain — Biological and Chemical, Cybersecurity, AI Self-improvement — name the THRESHOLD: what capability level is named (High, Critical, other), and who decides a model has approached it.
The Tradeoff the Framework Is Trying to Manage
Before judging the Framework as merely weak, see the tension it sits inside. A preparedness policy must balance two failure modes: deploying a dangerous capability too soon, and binding the organization so rigidly that legitimate development stalls or the policy gets quietly abandoned as unworkable.
Your reusable leak inventory
Carry this list into any commitment review.
- Designated-actor: Is any safeguard assigned to a role that can be reassigned, delegated, or vacated by the person it constrains?
- Unbound-overseer: For every oversight body named, is there a sentence that forces it to act on a specified trigger?
- Review-not-action: Does each threshold produce a constraint that holds, or only a discussion that can end in "proceed"?
A commitment leaks at its weakest binding, not its strongest promise checklist
- Designated-actor: Is any safeguard assigned to a role that can be reassigned, delegated, or vacated by the person it constrains?
- Unbound-overseer: For every oversight body named, is there a sentence that forces it to act on a specified trigger?
- Review-not-action: Does each threshold produce a constraint that holds, or only a discussion that can end in "proceed"?
- Conditional-escape: List every "as appropriate," "where feasible," "absent exceptional circumstances," "commercially reasonable"—and name who decides the exception applies.
- Stacking check: Do multiple leaks land on the same load-bearing safeguard? Stacked leaks compound; treat them as one critical failure, not several minor ones.
Map the Machine YourselfInside the book
Visual preview
A diagram you can keep open while you work.
Table of contents
10 chapters, built to be read in order.
01
The Enforcement Gap
02
Anatomy of a Preparedness Framework
03
Affordance Analysis for Safety Policy
04
Where Commitments Leak
05
If-Then Commitments That Bind
06
Designing Thresholds and Evaluations With Teeth
07
Governance Architecture for Accountability
08
Beyond Corporate Promises — The Government Backstop
09
Benchmarking Preparedness Maturity
10
Building a Preparedness Policy That Holds
158
Pages
17,539
Words
43
Exercises, checklists & tools
10
Chapters
Formats
Three formats. One purchase.
EPUB, PDF, and HTML are included so the book can work on an e-reader, as a designed copy, or as a searchable desk reference.
EPUB
For e-readers and reading apps.
The designed edition with diagrams and layouts intact.
HTML
Searchable, copy-pasteable, and practical as a reference.
Complete guide
The Preparedness Framework
$14.99
expert guide pricing
- ✓EPUB + PDF + HTML included in one purchase
- ✓10 chapters from the complete guide
- ✓DRM-free files for your own devices
- ✓7-day refund review for duplicate purchases, access issues, wrong files, or materially defective downloads
Secure checkout / instant download / tax handled at checkout
Before you buy
Questions, answered.
Does this include the full book?
Yes. You get the complete edition, including the chapter sequence and internal materials described on this page.
Which formats are included?
EPUB, PDF, and HTML are included so you can read on an e-reader, keep a designed copy, or use the searchable browser version.
What is the refund policy?
Because this is an instant digital download, broad change-of-mind refunds are not offered after the files have been accessed. Refund requests are reviewed within 7 days for duplicate purchases, accidental purchases before access, access failures we cannot fix, wrong files, corrupted files, or pages that materially misdescribe the book.