The Department of Homeland Security is pushing members of the critical infrastructure community to adopt practices aimed at ensuring the safe and secure use of artificial intelligence.
DHS today unveiled the new guidance, “Roles and Responsibilities Framework for Artificial Intelligence in Critical Infrastructure.” It proposes a series of voluntary responsibilities for the use of AI in the 16 U.S. critical infrastructure sectors.
The responsibilities are divided across five groups: cloud and compute infrastructure providers; AI developers; critical infrastructure owners and operators; civil society; and the public sector.
The detailed guidelines touch on a wide range of issues, including cloud environments, AI model and system design, data governance, deployment considerations, and the monitoring of AI use across critical infrastructure.
The framework was developed in conjunction with DHS’ AI Safety and Security Board. The board includes representatives from top AI companies, as well as executives of major computing and semiconductor firms, government representatives and civil society members.
Mayorkas said industry was “intensely engaged” in development of the framework.
“Industry was very, very helpful in ensuring that the guidelines are practical,” Mayorkas told reporters today. “This is not a document that advances theories. This is a document that provides practical guidance that can and should be implemented to advance safety and security.”
The framework comes after Cybersecurity and Infrastructure Security Agency identified three major categories of AI-related risks to critical infrastructure: attacks using AI, attacks targeting the use of AI, and design and implementation failures.
“For owners and operators of critical infrastructure whose essential services and functions the public depends on daily, understanding the nature of these vulnerabilities and addressing them accordingly is not merely an operational requirement but a national imperative,” the document states.
Public sector AI guidance
For government agencies in the “public sector” group, the framework recommends they “ensure that relevant private sector entities across all sectors are appropriately protecting the rights of individuals and communities, as well as a responsibility to respond and support the American public in times of crisis or emergency.”
For instance, the framework suggests agencies have the opportunity to use law and regulation to advance AI standards. “Laws and regulations should protect individuals’ fundamental rights, help drive innovation, advance the harmonization of different legal requirements, simplify compliance, and clarify incident reporting processes,” the document states.
At the same time, the framework also recommends the public sector “responsibly leverage AI to improve the functioning of critical infrastructure.”
“It should prioritize the development of, and funding for, programs that advance responsible AI practices in government services,” the document continues. “Public sector entities should engage with civil society and each other regarding the public sector’s use of AI and avoid using AI in a manner that produces discriminatory outcomes, infringes upon personal privacy, or violates other legal rights. Public sector entities should not fund discriminatory technologies.”
Outlook under Trump
However, the future use of the AI guidance within the federal government is uncertain. It was developed as part of President Joe Biden’s sweeping AI executive order, and President-elect Donald Trump has said he would repeal the EO.
“I of course cannot speak to the incoming administration’s approach to the board that we have assembled,” Mayorkas said. “I certainly hope it persists.”
But he said the framework “will endure,” adding that all 23 members of the AI Safety and Security Board support the practices.
“We expect the board members to implement the guidelines, to catalyze other organizations in their respective spheres and across the ecosystem, to adopt and implement the guidelines as well, and to have this take hold and to become the framework that will assist in in driving harmonization, which is so key to our leadership,” Mayorkas said.
DHS AI use cases
DHS has also been touting its own use of AI under several pilot projects.
“Our pilot projects have demonstrated tremendous AI capabilities to advance our mission, and so we are taking those pilots and actually integrating the AI successes into our operations,” Mayorkas said.
In late October, DHS announced the completion of three generative AI pilots.
As part of one project, U.S. Citizenship and Immigration Services used a GenAI tool to help train immigration officers on interacting with refugee and asylum seekers. DHS said the pilot was only used for training and not to make immigration eligibility determinations.
Based on the results of the USCIS pilot, DHS is examining how generative AI could be used for other training “as a supplemental tool to better prepare the next generation of DHS officers.”
Another DHS pilot project involved Homeland Security Investigations using large language models to produce summaries of law enforcement reports.
“The pilot showed that these were valuable tools to enhance investigative processes,” DHS announced in its press release. “The HSI pilot, which was developed using an open-source AI model, found that open-source models provided the flexibility necessary to experiment and measure effectiveness. HSI professionals continue to test and optimize the use of open-source models in supporting law enforcement investigations.”
Meanwhile, the third completed pilot involved the Federal Emergency Management Agency using an LLM to help state and local communities draft community resilience plans.
“FEMA learned that increasing user understanding of AI and receiving feedback directly from community users is an important first step to integrating GenAI into any existing process,” DHS said. “FEMA is using lessons learned from the pilot to help determine how the technology can best support their mission.”
Copyright
© 2024 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.