Preparedness Framework
In plain wordsOpenAI's published process for testing its AI models for catastrophic risks — such as help with biological weapons or cyberattacks — and deciding what safeguards are needed before release.
OpenAI's framework for tracking and mitigating catastrophic risks from its frontier models. The December 2023 beta rated each model on a four-step scale (low, medium, high, critical) in four areas — model autonomy, persuasion, cybersecurity, and chemical, biological, radiological and nuclear threats; a model still rated high after mitigations could not be deployed, and one rated critical could not be developed further. Version 2 (April 2025) narrowed the tracked categories to biological and chemical capability, cybersecurity and AI self-improvement; moved long-range autonomy, sandbaggingsandbagging, n.: When an AI deliberately performs worse than it can — for example on a safety test — so that it seems less capable or less dangerous than it really is., autonomous replication and undermining safeguards into research categories; dropped persuasion; and set two thresholdscapability threshold, n.: A level of dangerous ability — such as meaningfully helping someone make a biological weapon — that, once an AI model reaches it, obliges its developer to add stronger safeguards or hold the model back.: reaching High means safeguards must be in place before a model ships, and reaching Critical means they are needed while it is still being built. Capability and safeguards reports go to an internal Safety Advisory Group, with leadership deciding. OpenAI first treated a model as High in biology with ChatGPT Agent in July 2025.
Under the Preparedness Framework, the model was treated as High capability in biology, so the bio safeguards shipped with it.
Origin
Announced by OpenAI on December 18, 2023, with a Preparedness team then led by Aleksander Mądry, three months after Anthropic's responsible scaling policyresponsible scaling policy, n.: A public rulebook in which an AI company promises to test each new model for dangerous abilities and to add stronger safeguards, or slow down, as its models become more capable.; the second version appeared on April 15, 2025.
Usage note
Version 2 said OpenAI might adjust its requirements if a rival released a high-risk system without comparable safeguards, a clause critics saw as inviting a race to the bottom.
Related
- responsible scaling policyresponsible scaling policy, n.: A public rulebook in which an AI company promises to test each new model for dangerous abilities and to add stronger safeguards, or slow down, as its models become more capable.
- Frontier Safety FrameworkFrontier Safety Framework, prop. n.: Google DeepMind's published plan for spotting when its AI models approach dangerous abilities and applying stronger security and safeguards before that happens.
- capability thresholdcapability threshold, n.: A level of dangerous ability — such as meaningfully helping someone make a biological weapon — that, once an AI model reaches it, obliges its developer to add stronger safeguards or hold the model back.
- if-then commitmentif-then commitment, n.: A promise by an AI developer of the form: if our model can do a particular dangerous thing, then we will have these protections in place — or we will stop.
- dangerous capability evaluationdangerous capability evaluation, n.: A test of whether an AI system can do things that could cause severe harm in the wrong hands or on its own — such as helping make a bioweapon, hacking computers or copying itself.
- system cardsystem card, n.: A public report released with an AI model that explains what it can do, how it was tested for safety, what risks were found and what protections were put in place.
Sources
- OpenAI (2025). Preparedness Framework, Version 2. April 15, 2025.
Mentioned in
OpenAIOpenAI, prop. n.: The American company that makes ChatGPT and the GPT models. It began in 2015 as a nonprofit research lab and is now a for-profit company overseen by a nonprofit foundation., capability thresholdcapability threshold, n.: A level of dangerous ability — such as meaningfully helping someone make a biological weapon — that, once an AI model reaches it, obliges its developer to add stronger safeguards or hold the model back., CBRN riskCBRN risk, n.: The danger that AI could help someone make or use chemical, biological, radiological or nuclear weapons — for example, by explaining hard technical steps an attacker could not work out alone., Frontier AI Safety CommitmentsFrontier AI Safety Commitments, prop. n.: Promises made in 2024 by major AI companies to publish safety plans for their most advanced models, including the danger points at which they would stop if the risks could not be controlled., frontier labfrontier lab, n.: One of the handful of companies building the most advanced AI models of the moment, spending billions of dollars on computing power to train them., gpt-ossgpt-oss, prop. n.: Two free-to-download reasoning models released by OpenAI in August 2025 — its first downloadable language models since 2019 — that anyone may run, modify and use commercially on their own hardware., responsible scaling policyresponsible scaling policy, n.: A public rulebook in which an AI company promises to test each new model for dangerous abilities and to add stronger safeguards, or slow down, as its models become more capable., AI Safety LevelAI Safety Level, n.: Anthropic's numbered scale, modeled on laboratory biosafety levels, for matching how dangerous an AI model could be with how strong the protections around it must be., Frontier Safety FrameworkFrontier Safety Framework, prop. n.: Google DeepMind's published plan for spotting when its AI models approach dangerous abilities and applying stronger security and safeguards before that happens., if-then commitmentif-then commitment, n.: A promise by an AI developer of the form: if our model can do a particular dangerous thing, then we will have these protections in place — or we will stop., MLE-benchMLE-bench, prop. n.: A test that gives AI agents 75 real data-science competitions from Kaggle and checks whether their solutions would have won medals against human teams.