The Week the Government Got Serious About Who Owns Your AI

Four stories from this week that matter if you work at the intersection of federal IT, acquisition, and national security.

I usually run this on Fridays. But this week didn’t wait.

Four things happened in federal IT over the last 72 hours that are directly relevant to how government buys technology, who gets to compete for it, and what we’re actually building when we build AI systems for federal use. If you work in this space, you need to know about all four of them.

Let’s go.

Story 1

GSA’s draft AI procurement clause is the most consequential federal contracting document of 2026.

The General Services Administration published draft contract clause GSAR 552.239-7001 in early March. The comment period just closed. And if you haven’t read it, you should stop and do that before you respond to another AI-related RFI or RFP.

Here’s what it proposes: the government owns all data inputs, all outputs, and any custom model developments made during contract performance. Contractors cannot use that data for model training. And AI systems must be “American AI Systems,” a term the clause doesn’t define.

The clause also grants the government the right to use an AI system for “any lawful government purpose,” and stipulates that an AI system must not refuse to produce data outputs based on the contractor’s or service provider’s discretionary policies. That last part is the one that’s generating real heat in the contracting community, and it should. It’s a direct response to the Anthropic-DoD dispute from earlier this year, where Anthropic refused to allow its products to be used for surveillance of Americans or lethal autonomous weapons. The government’s answer, embedded in this clause, is to write contracts that remove the vendor’s right to refuse anything legal.

Industry has pushed back. TechNet, NetChoice, CDT, Americans for Responsible Innovation, and a dozen law firms have all filed comments flagging versions of the same problem: the clause is so broad, and so vague, that compliance may be harder to demonstrate than noncompliance is to claim.

What I keep coming back to is this: the clause isn’t wrong to want accountability from AI vendors. It’s wrong in the way it tries to create it. Broad data ownership mandates and undefined “American AI” requirements don’t build a better AI procurement ecosystem. They build a smaller one, where the companies willing to accept the risk are not necessarily the ones building the best systems.

For contractors holding GSA MAS schedules, the clock is already running. Once a mass modification issues, you have 60 days to accept. Get your AI asset inventory started now.

Story 2

The White House is asking for a record $75.7 billion in civilian agency IT spending for FY2027.

That’s a $7.7 billion increase over current spending. VA is asking for a 62% increase. DOJ is up 40.5%. Treasury is up 48%.

I’ll be honest: the headline number is less interesting to me than the DOJ figure. A $4.3 billion IT request with a $111 million increase specifically for Justice Information Sharing Technology is a signal that law enforcement modernization is no longer being treated as a backburner conversation. For anyone building in that space, that matters.

The macro story is straightforward. The administration is spending on technology. The harder question is whether the acquisition frameworks can absorb the investment without creating more technical debt in the process. Based on what I’m seeing from OMB’s M-26-10, that’s exactly the question they’re trying to answer.

Story 3

OMB’s M-26-10 is putting CIOs in the spotlight starting in May.

Beginning next month, CIOs at major federal agencies have to submit monthly reports to OMB on approved IT contracts, including utilization rates and prices paid. The goal is to surface waste, duplication, and underused tools before they become line items on a GAO report.

Mark Forman, who helped write Clinger-Cohen, said something this week that I think is exactly right: the problem isn’t modernizing technology, it’s modernizing agency operations. You can collect all the data you want, but without a target architecture and a real business case, you just have more data sitting in a system nobody’s reading.

The practical implication for contractors: CIOs are now accountable for what’s on contract in a way they weren’t before. Vendors who tie their work to outcomes, make reporting straightforward, and speak in mission terms rather than product terms are going to find that relationship easier to maintain. That’s always been true. It’s more true now.

Story 4

NIST is building security standards for AI agents. This is the right call.

NIST announced this week it’s launching initiatives to define security standards specifically for AI agents, systems that autonomously take actions via APIs. Not AI tools that answer questions. AI that actually does things.

This is where the risk calculus gets more interesting. An AI system that retrieves information and summarizes it has a certain risk profile. An AI system that submits a form, queries a database, modifies a record, or executes a workflow has a fundamentally different one. The attack surface isn’t just the model; it’s everything the model can reach.

Federal agencies are moving faster on agentic AI than their governance frameworks can accommodate. NIST working on this is not a reason to wait. It’s a reason to build in a way that will be compatible with what’s coming, because it’s coming soon.

One more thing

A suspected Chinese breach of an FBI system exposed surveillance targets’ phone numbers this week. I’m not going to speculate on operational details. But I’ll say what I’ve said in every DHS and law enforcement IT conversation I’ve had for the past several years: Zero Trust is not a framework anymore. It’s a survival requirement. And the gap between where most federal systems are and where they need to be is not closing fast enough.

That’s what had my attention this week. It’s a lot. But it’s also exactly why this work matters.

See You at the Edge,

Bill

Editor’s Note:

This article was originally published on Substack and is republished here with permission from Bill McKenna. You can read the original post here: https://bmac4fedit.substack.com/p/the-week-the-government-got-serious

About Trilogy Innovations, Inc.

Trilogy Innovations, Inc., located in Bridgeport and Morgantown, West Virginia, is an SBA Certified 8(a) small and minority-owned systems and software engineering company that delivers superior technical solutions for a variety of industries across private and public sectors. By applying a strong work ethic and an unwavering commitment to excellence, Trilogy’s highly skilled and talented personnel have successfully applied these core values across a multitude of government agencies and businesses. Inc. Magazine named Trilogy to the Inc. 5000 for four consecutive years from 2022-2025. The company was also named a 2025 Elev8 GovCon Honoree by Orange Slices for demonstrating a culture of excellence for talent, partners, and clients, marking the second consecutive year Trilogy has received this recognition.

For more information about Trilogy, go to www.trilogyit.com.