Though Congress has yet to make a serious and comprehensive effort to take up the issue, the Biden administration has been chipping away at AI rules by fiat (and via non-mandatory guidelines) since 2022. First with the “Blueprint for an AI Bill of Rights” as ChatGPT was breaking onto the scene, to the administration’s executive order to federal agencies a year later in 2023.
A component of that executive order is now coming to fruition as the Office of Management and Budget (OMB) has finalized its requirements for federal agencies in overseeing and using their AI systems. Compliance with the AI rules has been scheduled for dates throughout 2024, with full safety and transparency requirements in place by December. The administration has also committed to something of a hiring blitz to support what appears to be a complex transition for at least some of the impacted agencies, to include the appointment of Chief AI Officers throughout federal departments.
New transparency and safety requirements for federal agencies accompanied by more manpower
The OMB was first directed to develop these AI rules as part of the October executive order, and had them in a public comment period until recently. The rules do not necessarily provide specific direction to federal agencies, however, and they will be expected to develop plans and policies on what is a pretty tight schedule.
This is in part being addressed by direction to federal agencies to step up hiring, with the Biden administration committing to hiring 100 new AI specialists by the end of the summer tasked with facilitating and coordinating implementation across the government. The OMB AI rules also specify that key federal agencies must appoint Chief AI Officers and convene AI Governance Boards, with these requirements among the earliest to kick in starting with a May 2024 deadline.
The government appears to be trying to avoid throttling innovation, and it is also working within constraints of what can be accomplished without Congressional involvement. To that end, much of the focus of the AI rules is put on systems that federal agencies use in direct interaction with the public and that might produce biased or harmful results if not overseen carefully.
AI rules to be in place by beginning of December
Federal agencies will be required to review AI systems and apply safeguards by December 1, and those that cannot be properly secured and guaranteed to be free of bias will have to be taken offline unless the consequences of their removal would create a greater security risk.
The AI rules require federal agencies to not just conduct ongoing testing and audits, but also to make much of their internal workings transparent to the public. That will include annual inventories of AI use cases, data used to train and support models, and also code when it can be done without creating a security risk.
Federal agencies will likely spend 2024 feverishly looking for qualified AI candidates and upskilling existing staff, as the 24 CFO Act agencies get their AI governance leadership in place over the next month.
