Skip to main content
OpenAI

Safety

Safety at every step

We believe in AI’s potential to make life better for everyone, which means making it safe for everyone

Teach

We start by teaching our AI right from wrong, filtering harmful content and responding with empathy.

Test

We conduct internal evaluations and work with experts to test real-world scenarios, enhancing our safeguards.

Share

We use real-world feedback to help make our AI safer and more helpful.

Safety doesn’t stop

Building safe AI isn’t one and done. Every day is a chance to make things better. And every step helps anticipate, evaluate, and prevent risk.

How we think about safety and alignment

Learn more

Protecting people where it matters most

We work with industry leaders and policymakers to reduce harm and protect people across critical areas.

Illustration of two people avatar, with one smaller than the other, depicting a parent and child relationship.

Child safety

Illustration of a pair of sunglasses and a hat, depicting the concept of anonymity and privacy.

Private information

An illustration of a human avatar inside of a shield depicting safety.

Deep fakes

An abstract illustration featuring a tilted balance scale and directional arrows, representing the concept of bias in decision-making or evaluation.

Bias

A stylized illustration depicting a ballot box with a check-marked paper being inserted, symbolizing the voting process in an election.

Elections

Learn more

Parental-Controls_Art Card 1080x1080
Introducing parental controls

Product

PFv2 artcard 1.1
Our updated Preparedness Framework

Publication

Update on disrupting > Cover
An update on disrupting deceptive uses of AI

Safety

safety-blog-cover-02
OpenAI safety practices

Safety

Go deeper on safety

Explore the safety evaluations hub

Sora 2 System Card

Sora 2’s advanced capabilities require consideration of new potential risks, including nonconsensual use of likeness or misleading generations. To address these, we worked with internal red teamers to identify new challenges and inform corresponding mitigations. We’re taking an iterative approach to safety, focusing on areas where context is especially important or where risks are still emerging and are not fully understood.
Learn more

Addendum to GPT-5 system card: GPT-5-Codex

This addendum outlines the comprehensive safety measures implemented for GPT‑5-Codex. It details both model-level mitigations, such as specialized safety training for harmful tasks and prompt injections, and product-level mitigations like agent sandboxing and configurable network access.
Learn more

GPT-5 System Card

In this system card, we label the fast, high-throughput models as gpt-5-main and gpt-5-main-mini, and the thinking models as gpt-5-thinking and gpt-5-thinking-mini. In the API, we provide direct access to the thinking model, its mini version, and an even smaller and faster nano version of the thinking model, made for developers (gpt-5-thinking-nano).
Learn more

Addendum to OpenAI o3 and o4-mini system card: OpenAI o3 Operator

This addendum to o3 and o4-mini system card describes the release of Operator, a Computer Using Agent capable of interacting with the web through its own browser. Operator is now powered by a o3-based model (instead of GPT-4o) with enhanced safety fine-tuning for computer use.
Learn more

Addendum to OpenAI o3 and o4-mini system card: Codex

This addendum to o3 and o4-mini system card describes Codex, a cloud-based coding agent powered by codex-1, a fine-tuned version of OpenAI’s o3 model tailored specifically for software engineering tasks.
Learn more

OpenAI o3 and o4-mini System Card

This is the first launch and system card to be released under Version 2 of our Preparedness Framework. OpenAI’s Safety Advisory Group (SAG) reviewed the results of our Preparedness evaluations and determined that OpenAI o3 and o4-mini do not reach the High threshold in any of our three Tracked Categories: Biological and Chemical Capability, Cybersecurity, and AI Self-improvement.
Learn more

Addendum to GPT-4o System Card: 4o image generation

This addendum to the GPT‑4o system card describes the marginal risks we’ve focused on, and the work we have done to address them.
Learn more

GPT-4.5 System Card

This system card outlines how we built and trained GPT‑4.5, evaluated its capabilities, and strengthened safety, following OpenAI’s safety process and Preparedness Framework.
Learn more

Deep research System Card

This report outlines the safety work carried out prior to releasing deep research including external red teaming, frontier risk evaluations according to our Preparedness Framework, and an overview of the mitigations we built in to address key risk areas.
Learn more

OpenAI o3-mini System Card

This report outlines the safety work carried out for the OpenAI o3-mini model, including safety evaluations, external red teaming, and Preparedness Framework evaluations.
Learn more

Operator System Card

This report outlines the safety work carried out prior to releasing Operator including external red teaming, frontier risk evaluations according to our Preparedness Framework, and an overview of the mitigations we built in to address key risk areas.
Learn more

Sora System Card

This system card outlines the resulting mitigation stack, external red teaming efforts, evaluations, and ongoing research to refine these safeguards further.
Learn more

OpenAI o1 System Card

This report outlines the safety work carried out prior to releasing OpenAI o1 and o1-mini, including external red teaming and frontier risk evaluations according to our Preparedness Framework.
Learn more

GPT-4o System Card

This system card takes a detailed look at speech-to-speech while also evaluating text and image capabilities.
Learn more