President Joe Biden has issued an executive order aimed at setting safety and security standards for artificial intelligence. It requires developers of the most powerful AI systems to share safety test results and related data with the U.S. government before making those systems public.
The order gives federal agencies new work on testing, infrastructure risks and cybersecurity. Its reach also touches civil rights, privacy and workplace concerns, but the role of recommendations—and the need for further legislation—leaves open how much force the measures will have.
Safety checks before public release
The order focuses on foundation models that could pose risks to national security, economic security or public health. Developers of the most powerful systems must notify the federal government and provide results from safety tests before deployment to the public.
That requirement makes government access to test information a central part of the policy. The order says the goal is to make AI systems safe, secure and trustworthy before companies release them.
The National Institute of Standards and Technology, or NIST, is tasked with developing standards for extensive red-team testing. The order also assigns the Departments of Energy and Homeland Security work on risks involving AI and critical infrastructure.
Agencies are asked to address specific risks
The executive order directs work across several areas where AI could create or intensify harm. These include the use of AI to engineer dangerous biological materials, fraud and deception powered by AI, and vulnerabilities in critical software.
It calls for a cybersecurity program to build AI tools that can help address weaknesses in critical software. The Departments of Energy and Homeland Security are expected to consider infrastructure risks as part of their responsibilities.
These measures connect AI oversight to existing government functions: setting technical standards, protecting infrastructure and improving cybersecurity. The order describes a broad effort to build safety considerations into AI development and deployment.
Civil rights and privacy remain open issues
The order also recognizes that AI can worsen discrimination and bias in areas such as healthcare, justice and housing. It raises concerns about workplace surveillance and job displacement, and calls for best practices on AI use throughout the criminal justice process.
Those practices would cover decisions and tools including sentencing, parole and probation, pretrial release and detention, risk assessments, surveillance, crime forecasting, predictive policing and forensic analysis. The order frames this work as guidance for fairness, rather than detailing a new enforcement system for those uses.
Privacy is another concern. The order notes that AI can make it easier to extract and exploit private data at scale, including in model training. It asks Congress to pass bipartisan data privacy legislation and calls for more federal support for privacy-preserving AI development techniques.
How much force will the order carry?
The executive order follows voluntary commitments secured by the Biden-Harris Administration from major AI developers, including OpenAI, Google, Microsoft, Meta and Amazon. Those commitments were intended as a step before the order, which now sets out additional government expectations.
Still, many elements appear to rely on recommendations and guidelines. The source article notes that it is unclear how enforceable the order will be without further legislative changes, particularly in areas such as data privacy and fair use in the justice system.
The order arrives amid wider efforts to shape AI governance. G7 leaders agreed on guiding principles and a voluntary code of conduct for developers through the Hiroshima AI Process; the United Nations announced a board to explore AI governance; and the U.K. was hosting a global summit on the subject at Bletchley Park.
Together, these efforts reflect the challenge governments face as generative AI advances: addressing safety, security and social harms while setting expectations for companies building powerful systems. The executive order establishes federal priorities, but its practical impact will depend on how agencies carry out their assignments and whether additional legislation follows.