Strengthening democratic oversight in national security

AI is changing how democratic governments protect their people. It can help stop cyberattacks, secure critical infrastructure, detect threats earlier, and give public servants a clearer picture in a crisis. Used well, these tools can strengthen national security.

Democratic oversight helps ensure that uses of public power remain accountable to the people they serve. As AI makes national security work faster and more capable, the institutions responsible for overseeing that work also need to keep pace. AI can act on missing or outdated context, misconfigured objectives, or misaligned assumptions at speed and scale, making mistakes harder to detect manually before they spread. Oversight is especially important in AI use for national security because decisions may necessarily involve secrecy and can have high-stakes consequences.

Guided by our National Security Principles, we approach national security deployments with safeguards, meaningful human judgment, and appropriate oversight, especially in high-stakes settings. Those principles also make it a priority to help the institutions charged with oversight build the capacity to oversee AI use at speed and scale.

Ensuring oversight keeps pace with AI adoption

Government institutions already have established authorities and responsibilities for overseeing national security activities. When AI contributes to those activities, oversight may need to evolve. Traditional oversight is labor intensive and becomes nearly impossible when small teams using paper-era methods are asked to review systems operating at machine speed.

OpenAI believes more capable systems require stronger oversight, and we apply that principle to ourselves. Under our Preparedness Framework, advanced models undergo rigorous review before deployment to understand their capabilities and determine whether safeguards are sufficient. That scrutiny continues after deployment through ongoing monitoring and incident review, with what we learn informing stronger safeguards and future model development.

OpenAI does not perform an oversight role over the government, nor should it. That responsibility rests with elected officials and the public institutions established for the purpose. Rather, we can support oversight institutions with our technology as they carry out their existing mandates.

Three principles will guide this work.

_First, AI should strengthen, not replace, human and institutional judgment._ Decisions about whether government conduct is lawful, appropriate, and consistent with applicable rules and policies belong to authorized agencies and legally constituted oversight bodies.

_Second, government use of AI should be traceable and legible to authorized oversight._ Reviewers should be able to understand how AI contributed to a decision while protecting classified, private, and security-sensitive information.

_Third, AI should be used to empower democratic oversight institutions._ Manual oversight of AI use in government cannot scale. As AI is adopted in national security and other high stakes government work, it should also be used by oversight bodies in a responsible and legible manner in accordance with mandates.

What OpenAI will do

Over the next year, we plan to advance this initiative in the following ways:

* We will work with authorized officials to identify practical opportunities where AI tools or technical support can make oversight more effective.

* We will provide $5 million in training, technical support, and OpenAI credits to democratic government oversight bodies to support responsible adoption and evaluation of AI tools.

* We will work with oversight institutions to pilot tools that help authorized reviewers examine the records surrounding AI-assisted government decisions including relevant inputs, outputs, and tool use, to better understand how AI affected outcomes. Tools will be interoperable or model-agnostic where feasible. Participating institutions, and not OpenAI, will retain control of the evidence, outputs, and findings, which will remain available only to authorized government officials.

* We will engage civil society and technical experts so their perspectives inform how OpenAI develops these tools and approaches the responsibilities that come with them.

The choice in front of us

AI can concentrate power, or it can help democratic institutions exercise and check power more effectively. Technology alone will not decide which future we get, but the choices human institutions and companies make now will.

We will judge this initiative’s progress by whether authorized reviewers are better able to fulfill their obligations and the public has stronger reasons to trust that government AI systems are being used as intended.

As government use of AI evolves, OpenAI will continue supporting the institutions responsible for its oversight.