DHS needs to beef up government’s AI risk assessments, watchdog says

The Department of Homeland Security should improve guidance for risk assessments focused on artificial intelligence and the threats the technology could pose to critical infrastructure sectors, the Government Accountability Office said in a report released Wednesday.

The watchdog report, which recommends that DHS “quickly” update its “guidance and template for AI risk assessments” to address various gaps, comes amid ongoing concern that AI could be used to undermine critical infrastructure sectors, which range from dams and IT systems to emergency services.

The GAO focused its report on how various sector risk management agencies (SMRAs) have addressed six activities tied to AI assessments: documented assessment methodology, identified AI use cases, identified potential risks, evaluated level of risk, identified mitigation strategies, and mapped mitigation strategies to risks.

While the SMRAs — agencies charged with protecting U.S. critical infrastructure sectors —  completed requirements to identify AI use cases, they didn’t fully consider the risks involved, according to the GAO. Sixteen of the 17 risk assessments analyzed by the watchdog considered potential risks, but none included “a measurement of both the magnitude of harm (level of impact) and the probability of an event occurring (likelihood of occurrence).”

While most agencies identified risk mitigation strategies, most didn’t connect those strategies to potential risks. Seven agencies partially addressed risk mitigations, while 10 didn’t do so at all. 

DHS concurred with the GAO’s recommendation that it “expeditiously” update its guidance “to address the gaps identified in this report, including activities such as identifying potential risks and evaluating the level of risk.” 

House AI task force releases final report, eyes future work with Trump administration

House AI task force members on Tuesday touted the long-awaited release of their report outlining recommendations to Congress on the federal government’s adoption of AI, while confirming that lawmakers are in conversation with the incoming administration about the very same topics.

The nearly 300-page report digs into a plethora of areas that the emerging technology touches, including the government’s own use of AI, along with recommendations concerning AI governance and the AI workforce.

Rep. Jay Obernolte, R-Calif., co-chairman of the task force, said during a press conference Tuesday that legislators and the incoming Trump administration are discussing what could be done to the executive order that the Biden White House released last fall, what items need legislative action and how efforts might be “harmonized” between the legislative and executive branches. 

“I am very happy and impressed that the incoming administration is paying so much attention to this issue,” Obernolte told FedScoop after the press conference. “I’m delighted that they’ve named an AI czar — that indicated that we share the belief that AI is going to be transformative and that we share our belief of the need to engage this topic.”

Those comments echo similar sentiments expressed by another task force member,  Rep. Kat Cammack, in an interview with FedScoop weeks ago. The Florida Republican said she expects the incoming administration to produce a framework that could look to ensure that large language models aren’t producing unintended results. Cammack called the Biden administration’s approach to the technology “more heavy-handed than I would have liked to have seen,” adding that “all indications from the Trump team, and the people that they’re building in their cabinet, it really reflects a more industry-driven approach.” 

“We can’t keep up with how fast the AI industry and the technology is evolving, so it’s better for us to be on the side of … [letting] industry do what they do best, and that’s innovate,” Cammack told FedScoop.

Obernolte confirmed that the task force met with President-elect Donald Trump’s technology transition team last week, and that conversations with the newly appointed AI czar were set to happen this week.

The report, Obernolte said, is “only the beginning” of the AI conversation and will only be effective when future Congresses implement “some of the recommendations that we have created.” 

The task force recommended that agency AI policies be required to include “plain language notification and appeals processes” for individuals or organizations that are impacted by determinations that are “substantively and meaningfully augmented by an AI system.” Additionally, the document suggests that lawmakers consider existing laws and policies that pertain to federal IT to avoid inefficiency, confusion, competing requirements and more.

Significantly, the final report also embraces the need to incentivize AI workers to join the federal workforce through the creation of an AI scholarship for service program and “continuing to support the development of AI roles in federal agencies.”

Rep. Bill Foster, D-Ill., told FedScoop that the recommendations in the final draft are mostly about “flagging issues that Congress may want to consider getting involved in.” Foster said that earlier draft versions of the report were “more restrictive in singling out individual pieces of legislation.”

Foster argued for making sure that AI developments and decisions are being made with “a very strong human in the loop,” pointing specifically to the commercial sector’s use of the technology in consumer-facing interactions, like call centers or customer support. Maintaining a consumer-facing operation, he said, is “one of the strongest efficiencies that you get from advanced AI in this area, and I would be disappointed if some of that efficiency wasn’t captured by the federal government.”

As the non-agency Department of Government Efficiency looks to reduce federal government waste through the workforce, Foster said the task force is focused on a bipartisan approach to identifying “common ground.”

“Many of the things that are listed publicly or tweeted out by the members of the DOGE commission … are things that could receive bipartisan support if they’re structured properly,” he said.

CISA delivers new directive to agencies on securing cloud environments

Adele Merritt named CIO of National Institutes of Health

Adele Merritt has been named the top IT official for the National Institutes of Health, giving the agency a permanent chief information officer for the first time in roughly two years.

Merritt, who was most recently the CIO of the intelligence community, began her roles as CIO and director of the agency’s Office of the Chief Information Officer on Monday, NIH Director Monica M. Bertagnolli said in a written announcement. 

According to Bertagnolli, Merritt comes to NIH with more than 20 years experience in cyber and national security operations and has held positions across the government intelligence community. Merritt, a mathematician, was also previously the Department of Energy’s principal deputy CIO for cyber and held senior positions on the staff of the White House National Security Council, according to the announcement.

“Her breadth of experience and passion for harnessing technology to drive mission outcomes will be invaluable to NIH in advancing cutting-edge IT solutions in support of biomedical research and public health,” Bertagnolli said.

Merritt takes over for Dennis Papula, who served as the acting CIO and acting director of the OCIO since January 2023, following the retirement of NIH’s longtime CIO Andrea Norris. Papula similarly has a new role as the new CIO of the Administration for Strategic Preparedness and Response, which is also part of the Department of Health and Human Services. He started that role Dec. 15. 

The announcement comes a few weeks after HHS named Jennifer Wendel as its permanent CIO. Wendel had been serving as acting CIO since Karl Mathias departed the agency for a role at NASA nearly a year ago. 

Report highlights how secure data-sharing platforms can support the Intelligence Community’s IT roadmap

Department of Homeland Security reveals nearly 160 ways it’s using AI 

The Department of Homeland Security on Tuesday released the latest version of its artificial intelligence use case inventory, reporting 158 active applications for the technology — a major jump from the 67 it made public last year. 

In an explanatory blog post outlining how the agency approaches artificial intelligence, Eric Hysen, the agency’s CAIO and CIO, said that 29 deployed AI use cases and 10 upcoming AI use cases were deemed to be rights- or safety-impacting, a new level of scrutiny established by recent White House guidance. Roughly half of those deployed use cases included technologies related to face recognition and face-capture technologies, he said. 

The use cases for Customs and Border Protection include technology from the company Babel, a passive body scanner meant to help with weapons detection at pedestrian border crossings and an autonomous underwater vehicle. Immigration and Customs Enforcement is using a facial recognition-based biometric check-in tool, as well as facial recognition technology for identifying victims of child exploitation and a facial recognition service used in other investigations. The Transportation Security Administration has listed AI-enabled Axon body cameras and a generative AI system for employee workflows. More details about these use cases are available on the inventory. 

The document also mentions an internal generative AI chatbot used by the Cybersecurity and Infrastructure Security Agency.

Hysen said in the blog posted that he “reviewed and approved each use case to ensure they were meeting each practice, including testing performance in a real-world context, maintaining human oversight and accountability, conducting ongoing monitoring and mitigation for AI-enabled discrimination, among other requirements.” He noted that he did not need to issue any waivers of the required risk management practices. The Office of Management and Budget approved five short-term compliance deadline extensions for five safety- or rights-impacting use cases, he added. 

DHS said its goal with the release was to expand the inventory and err on the side of transparency, a decision based on guidance from the agency’s governing board to reveal as much information as possible, “even if some details about certain use cases cannot be publicly shared.” AI related to the intelligence community is excluded, based on a recent national security memorandum, but the agency will release more information about approaching those use cases in April. 

The full inventory, which also includes deactivated use cases, is available here. 

FedScoop has continued to report on the accuracy and transparency of AI inventories, including changes to DHS’s inventory made last year. Researchers at Stanford have also tracked these inventories, and a government watchdog previously flagged issues with DHS’s tracking.  

ChatGPT, meet DHSChat: Homeland Security has a new AI bot

The Department of Homeland Security has built an artificial intelligence-based chatbot it’s calling DHSChat, the latest in a string of agency experimentations with generative AI.

DHSChat is designed as a chatbot meant for internal use within the agency, according to a press release shared Tuesday. The tool was developed by DHS’s AI Corps and its director, Michael Boyce, after employees experimented with commercial generative AI tools, including Claude and ChatGPT. 

In a statement, Homeland Security Secretary Alejandro Mayorkas said that the “cutting-edge tool” would help workers across DHS “draft vital reports, summarize critical information, develop new software, streamline administrative tasks, and much more. Our Department is committed to embracing innovative new technologies like GenAI in a safe, secure, and responsible way, all while maintaining the highest standards of security and protecting civil rights, civil liberties, and privacy.” 

The tool is similar to commercial tools but was made internally and operates in a secure environment, Boyce wrote in a blog post. In response to a FedScoop question, a spokesperson for the agency said that the chatbot leverages external large language models through an API. “Data from DHSChat is not used to train external models,” the spokesperson added.

The tool is now available to more than 19,000 workers at the agency’s headquarters, as well as in pilot form to 10 operating agencies. 

According to privacy paperwork published by DHS last year, ChatGPT, BingChat, Claude 2, DALL-E2, and Grammarly all had provisional approval within the agency. DHSChat is meant to follow up on an AI roadmap released by the agency earlier this year. Use cases for the agency include assisting with cyber terminology and creating study aids, per the blog post.

“Last year, we began allowing employees to use commercial generative AI tools like ChatGPT and Claude on publicly available information only; thousands of employees have now received training and use these tools to support their work,” Boyce wrote.

He continued: “By using DHSChat, employees are able to perform routine work more efficiently. This includes summarizing complex documents and reports, generating computer code, and streamlining repetitive tasks like data entry. With this new tool, thousands of employees will be able to leverage generative AI capabilities safely and securely using non-public data.”

Eventually, Boyce added, the agency hopes to create a “secure internal knowledge hub” that employees can query. 

The announcement comes as agencies begin to release the latest versions of their AI inventories, including DHS.

Correction: An earlier version of this article referred to ChatDHS, rather than DHSChat.

HHS reports 66% increase in AI uses as agencies release inventories

The Department of Health and Human Services was among the first federal agencies to release its 2024 AI use case inventory Monday, reporting a roughly 66% increase in uses from the previous year.

In a post accompanying the inventory, Steven Posnack, HHS’s principal deputy assistant secretary for technology policy and principal deputy national coordinator for health IT, highlighted the health agency’s increase from 163 to 271 use cases and noted the varying stages of development. The new figure, for example, includes uses for operation and maintenance and in acquisition and development, as well as 16 uses that are retired.

The inventory comes as agencies across the federal government are expected to release new and expanded lists of AI use cases for 2024. Agencies were first required to release annual and public AI use case inventories under a Trump-era executive order, but the reporting has been inconsistent and in some cases even incorrect, as documented by Stanford researchers, FedScoop reporting, and a government watchdog review.

This year’s inventories, which were due Monday, were anticipated to be more expansive than years past following a memorandum from the Office of Management and Budget and instructions that added additional information to and clarified the existing process. 

In addition to HHS, at least the Department of Veterans Affairs, Department of Agriculture, Environmental Protection Agency, Social Security Administration, Department of Homeland Security, and Department of Housing and Urban Development have also released their 2024 inventories. All of those agencies similarly reported more use cases than the previous year per their totals on a consolidated list of all agency inventories released by the White House in October 2023.

Artificial intelligence has been a focus for President Joe Biden, who released an executive order last year aimed at using the technology while ensuring its safety and trustworthiness. That effort carried into the inventory process through the introduction of specific risk management practices for use cases that could impact Americans’ rights and safety. 

Per the OMB memo, all rights- or safety-impacting AI had to be brought into compliance with the specific risk management practices — such as assessing the risks and data quality — by Dec. 1. If they weren’t in compliance, and the agency hadn’t received approval on a year-long extension, those uses had to be halted.

Agencies also must disclose whether a use case is rights- or safety-impacting on their inventories. HHS, for example, listed two rights-impacting and two safety-impacting uses. The two rights-impacting uses by the Health Resources and Services Administration are tools to support evaluation of scholarship applications for medical students. Both use cases were in the “initiated” phase. 

Meanwhile, the two safety-impacting uses focused on data about family separations and children who have left the care of the Office of Refugee Resettlement. Both uses were under the Administration for Children and Families and were in an acquisition and/or development phase.

HUD, for its part, reported a use case that was both rights- and safety- impacting. That use case deploys Google Translate on its public-facing website to provide information in other languages. It is currently in an operation and maintenance phase.

The White House didn’t immediately respond to a request for comment on findings and what to expect for the inventories this year.

NASA looking for new electronic health record system to support astronauts

NASA is researching options to replace an existing electronic health record system that supports some unique patients: astronauts.

In a request for information, the agency said it’s looking to replace a system that currently holds roughly 30,000 records for NASA employees that date back to the founding of the U.S. Space Program. The request, which is purely information-seeking and may be used to finalize a request for quotes, was initially posted in November and responses were due Friday.

While that current system, which is installed primarily at the Johnson Space Center in Houston, is understood well and configured for the agency’s current workflows, it’s limited in its “patched together work arounds,” “manual processes and procedures,” and “single points of failure in both system and human terms,” NASA said in response to questions on the request posted to SAM.gov Thursday. 

The agency also pointed to the lack of an integrated patient portal and challenges with the user interface as drawbacks.

A new system, the agency said, could lead to improvements such as automation of process, procedure, and workflow; better data management; “robust connectivity to and from NASA systems external to the EHR that are involved in astronaut care;” and an environment that’s “simpler” and “leaner.”

The Human Health & Performance Directorate uses the existing EHR as one of the products that supports its “operational decision-making,” according to the request. That directorate is the component within NASA focused on “health and productivity of humans during all phases of spaceflight,” including flight- and ground-based healthcare, research, and test subject screening.

Among the unique attributes of the system: NASA said it doesn’t bill or allow self-scheduling, and while they do collect insurance information, they don’t file claims. The system also relies heavily on custom reporting for data fields such as flight and dive qualifications and the requirements needed to maintain certification. 

Unique services of that system include “support for reacclimating astronauts physically, psychologically, and operationally when they return from microgravity” and “workflows to obtain data from astronauts in flight to support medical monitoring by physicians on the ground,” NASA said.

Some things work well with the current system. NASA noted the existing EHR has “limited cloud-based test environments and connectivity” in addition to backup facilities in White Sands, New Mexico. And, according to the request, that “configuration is stable, well-understood, expandable, secure, and highly functional.”

Per the documents, the current EHR is commercial-off-the-shelf and the replacement would also be commercial-off-the-shelf. Implementation of the new software would need to be completed no later than July of 2026.

FBI, DHS supporting New Jersey drone response with detection methods

The FBI and Department of Homeland Security disclosed Thursday that they are supporting New Jersey law enforcement with “numerous detection methods” as part of the response to recent mysterious aircraft sightings in the state’s skies.

Thus far, however, the federal agencies “have not corroborated any of the reported visual sightings with electronic detection,” according to a joint statement shared with FedScoop and DefenseScoop. 

“To the contrary, upon review of available imagery, it appears that many of the reported sightings are actually manned aircraft, operating lawfully. There are no reported or confirmed drone sightings in any restricted air space,” the FBI and DHS statement said.

The FBI declined to provide additional details on the capabilities of those technologies.

The agencies said they and federal partners are working closely with New Jersey State Police, and “continue to deploy personnel and technology to investigate this situation and confirm whether the reported drone flights are actually drones or are instead manned aircraft or otherwise inaccurate sightings.” 

The statement didn’t provide additional details on what other federal agencies were involved, but U.S. Northern Command, a component of the Defense Department focused on homeland protection and defense, confirmed to DefenseScoop that it is monitoring the situation.

The sightings of what people have thought to be drones in recent weeks have sparked furor and lawmaker demands for answers. They also come amid a broader conversation about policy and transparency surrounding government response to unidentified aerial phenomena, or UAP.

New Jersey and New York lawmakers, including Sen. Cory Booker, D-N.J., and Majority Leader Chuck Schumer, D-N.Y., wrote to leaders at DHS, the FBI and the Federal Aviation Administration earlier this week asking for a briefing on the response, citing public concerns.

Some residents, they wrote, have reported seeing drones larger than what is commercially available and others have been spotted near sensitive locations such as critical infrastructure and military installations. 

In addition to privacy, the lawmakers said they were also concerned about public safety, citing reports that “a medevac helicopter was prevented from transporting a seriously injured patient for care because of the presence of these drones.” They specifically cited a story from Popular Science on that instance.

Lawmakers also probed an FBI official about the New Jersey sightings at a House Homeland Security joint subcommittee hearing earlier this week.

The FBI and DHS said they take the threat of unmanned aircraft systems seriously, saying that’s “why law enforcement and other agencies continue to support New Jersey and investigate the reports.” 

While they emphasized that so far no malicious activity has been uncovered in the state, the agencies said “the reported sightings there do, however, highlight the insufficiency of current authorities.”

The FAA didn’t immediately respond to a request for comment.

DefenseScoop reporter Brandi Vincent contributed to this story.