Agencies Are Getting Good at Buying AI But Still Have Trouble Securing It

Federal agencies are getting better at buying advanced technologies like artificial intelligence but still lag in deploying those tools due to security concerns, according to a Homeland Security Department procurement official.

“We’re doing a really good job at procuring things,” Jessica Clark, an official on the Acquisition Systems Team in DHS’s Office of the Chief Procurement Officer, said Tuesday during the Professional Services Council’s annual Tech Trends conference. “But getting it up and running safely is always going to be an issue for our program managers.”

Clark said DHS takes a different procurement strategy when looking at new and innovative technologies, preferring a phased approach where a relatively large pool of vendors is whittled down over the course of multiple prototypes and demonstrations, with each subsequent phase using larger datasets that are more and more relevant to the program.

She cited the department’s work integrating AI into the Contractor Performance Assessment Reporting System, or CPARS, which contracting officers use to gauge a vendor’s past performance on government contracts. For the CPARS AI effort, DHS started with nine vendors, which was then down-selected to six and then four.

At that point, DHS partnered with the Health and Human Services Department to ensure the resulting set of contracts would be viable across government.

“Our goal is not just to get an AI, but to get multiple vendors that could provide AI that potentially we could then put on a [governmentwide acquisition contract] or some kind of best-in-class contract,” Clark said. “Then, agencies throughout the federal government would have a choice of vendors to come and look at to help with their CPARS—their contractor performance assessments.”

But DHS and other agencies are finding they have “a lot of flexibility” on the procurement side by using different acquisition methods that allow for this phased approach. However, getting those tools and services operating on real data in real federal environments has proved more challenging.

“Just because you can buy it doesn’t mean you can use it,” she said. “Our biggest problem at that point is handing it over to our program managers” who need to get an authority to operate, or ATO, which confirms the tool or service meets baseline security standards.

Getting third-party technologies accredited as secure has been an ongoing issue across government for years. In 2012, the Federal Risk and Authorization Management Program, or FedRAMP, was established to help with this process by granting provisional ATOs that agencies can use as a starting point.

But reusing ATOs from FedRAMP or other agencies—a method called reciprocity—has been a perennial issue in government, to the point where lawmakers have repeatedly introduced legislation that would require agencies to reuse authorizations.

“Just because you have a vendor that is potentially working at a different agency, that authority to operate is not a guarantee to move over to your agency,” Clark said, noting that there are multiple levels within an ATO based on how the technology will be used by the authorizing program.

“All of those things lead into not just, ‘Can I buy it? Finally, I found a solution that meets all of my requirements.’ But can I actually put it in place in a reasonable time with a reasonable cost—because ATOs do cost time and money,” she said.

While some agencies—including the Defense Department—have pushed to reuse ATOs as much as possible, that is often more difficult for cutting-edge technologies and highly sensitive programs and datasets, such as using AI in programs that handle confidential or personally identifiable information.

“The ATO process is not expected to get any shorter given the cybersecurity issues that have happened,” Clark said, noting recent breaches like Colonial Pipeline and SolarWinds. “It’s not going to get any easier, and it’s necessary for safe government.”