---
title: "Preparedness Framework: what it means · The AI Compute Race · DRYBURGH"
source: https://dryburgh.com/vol1/glossary/preparedness-framework
description: "Preparedness Framework: The level of cyber capability, in OpenAI's Preparedness Framework, at which a model can find and build working exploits across many…"
---

# Preparedness Framework

What does “Preparedness Framework” mean?

Preparedness FrameworkSee Critical.

[Critical](https://dryburgh.com/vol1/glossary/critical) The level of cyber capability, in OpenAI’s Preparedness Framework, at which a model can find and build working exploits across many well-protected systems without human help. The framework then requires safeguards during development as well as before release. *Chapter 3.*

From *The AI Compute Race*. [Read the book](https://dryburgh.com/vol1#buy)

Defined in [Chapter 3 — The Missing Researcher](https://dryburgh.com/vol1/notes/chapter-3).

## Sources that use it

- OpenAI, *Preparedness Framework*, version 2, 15 April 2025 (Critical cybersecurity threshold; safeguards required during development regardless of deployment intent; development halted until Critical-level control standards are specified and met). [In the Sources](https://dryburgh.com/vol1/sources/company#openai-preparedness-v2)
- Alex Heath, “OpenAI Is Slowing Down Its AI Training,” *TIME*, 18 August 2026 (Astra paused “a little more than two weeks”; largest planned frontier run on hold; Altman: “We’ve shifted a lot of compute, not just to alignment research, but also to these new monitoring systems”; possible “Critical” cyber classification under the Preparedness Framework; roughly a week to discover the breach; Anthropic’s weakened commitment recalled from TIME’s own February report, with Jared Kaplan’s “if competitors are blazing ahead” from that report; Pachocki on the lapse: “For AI, you should expect the unexpected”; Altman: “I think it is a good time to slow down”, TIME’s own scope in the same sentence being that he was “describing the company’s decision to halt the development of its most powerful, unreleased models”; TIME’s account of the lapse: OpenAI “had built monitors capable of inspecting what its models were planning, but had not applied them to the system in the evaluation because it underestimated their capabilities”). [In the Sources](https://dryburgh.com/vol1/sources/press#time-openai-slowdown)

## Related terms

- [Critical](https://dryburgh.com/vol1/glossary/critical)
- [Astra](https://dryburgh.com/vol1/glossary/astra)

From the glossary of The AI Compute Race.

- [Campuses](https://dryburgh.com/vol1/ledger)
- [Tests](https://dryburgh.com/vol1/tests)
- [Scorecard](https://dryburgh.com/vol1/scorecard)
- [Data](https://dryburgh.com/vol1/data/releases)
- [Chapter by chapter](https://dryburgh.com/vol1/companion)
- [The record at a glance](https://dryburgh.com/vol1/observatory)
- [The machine](https://dryburgh.com/vol1/machine)
- [One campus, four futures](https://dryburgh.com/vol1/hyperion)
- [Glossary](https://dryburgh.com/vol1/glossary)
- [People](https://dryburgh.com/vol1/people)
- [Sources](https://dryburgh.com/vol1/sources)
- [Notes](https://dryburgh.com/vol1/notes)
- [Where the book was wrong](https://dryburgh.com/vol1/corrections)
- [Where its sources were wrong](https://dryburgh.com/corrections?book=vol1)
- [The Paper](https://dryburgh.com/vol1/paper)
- [Consent](https://dryburgh.com/vol1/consent)
- [Who builds and pays](https://dryburgh.com/vol1/parties)

*The AI Compute Race: Five Races, One Machine* is a book by Lee S. Dryburgh, the first volume of THE COMPUTE ECONOMY, published by Dryburgh, an imprint of Contestra OÜ, Tallinn, Estonia; its record closes on 8 November 2026.

[About the book](https://dryburgh.com/vol1)[Tell me when it is out](https://dryburgh.com/vol1#notify)