
Illustration: Maura Losch/Axios
OMB on Thursday issued government-wide policy for mitigating AI risks, a major part of President Biden's AI executive order.
Why it matters: Agencies across the federal government now have specific guidance on how to use AI in their work, and how to ward off its risks.
- "President Biden and I intend that these domestic policies will serve as a model for global action," Vice President Kamala Harris said on a call with reporters.
As we noted in our deep dive Wednesday, this was one of the tasks due Thursday, 150 days since Biden issued his sweeping AI EO.
- This OMB memo is the final version after a draft was released in November.
- The White House said federal agencies completed all 150-day actions required in the EO.
What's inside: The guidance focuses on the privacy and safety of the public, transparency in processes and how AI can make government services easier to access, Harris said.
- By Dec. 1, agencies must "implement concrete safeguards when using AI in a way that could impact Americans' rights of safety," per a White House fact sheet; that includes being able to opt out of things like TSA facial recognition.
- Companies that want to contract with the federal government to sell their AI systems will have to meet certain specifications based on the guidance.
Transparency is a big focus. Agencies will have to release inventories and metrics of their AI use cases, along with making public government-owned AI code.
- The guidance also encourages agencies to use AI to try to solve problems around climate change, public health and transportation safety by lifting some bureaucratic barriers.
- It aggressively directs agencies to "upskill their AI talent" and includes a commitment to hire 100 AI professionals by this summer.
Our thought bubble: Agencies are going to be laboratories for how AI can be implemented in a major workforce touching all parts of the American economy.
- Congress may be stalling on AI legislation, but it's clear the Biden administration thinks it can make its mark on how AI should be used in government.
