Back to News

The Federal Government Is Buying More AI. Here's What We Believe Makes a “Strong” AI Vendor

The federal government has moved well beyond small artificial intelligence experiments. In April 2026, the Government Accountability Office reported that federal agencies more than doubled their use of AI from 2023 to 2024 and continued acquiring additional AI capabilities through fiscal year 2025. The White House released its “AI Action Plan” last year to outline the Administration's strategy for strengthening American leadership in artificial intelligence.

That growth and planning create opportunities for technology companies. It also raises an important question for vendors pursuing federal work:

What sets one vendor apart from another in today’s world of AI?

To answer this question at Kleos Technology, we have carefully developed a core set of principles we strive to follow when designing AI systems for government customers, and when considering whether or not we possess the capability necessary to bid on certain contracts. They are informed by current federal guidance, our own engineering philosophy and experience, and our belief that reliable AI requires far more than simply connecting an application to a language model.

Start With The Mission

The conversation should begin with a clear understanding of the operational need, and OMB guidance encourages agencies to use performance measures that connect proposed AI capabilities to actual mission outcomes. We believe a vendor should consider the problem being addressed, who will use the system, what decisions it will support, and what could happen if it produces a wrong result.

AI is not the best answer to every requirement. Some problems are better addressed through traditional software, structured rules, improved search, workflow automation, or a combination of several technologies. The value comes from being able to match available technologies to the specific mission in the most efficient and effective manner.

Provide Evidence Beyond The Demonstration

AI demonstrations can be persuasive. They can also hide weaknesses and lead to a false sense of confidence.

Any vendor can choose a favorable prompt, use clean data, and present a controlled scenario that produces an excellent result. That does not account for real-world scenarios where information is incomplete, ambiguous, outdated, or unfamiliar. As Prussian Field Marshal Helmuth von Moltke famously stated: "No plan of operations reaches with certainty beyond the first encounter with the enemy."

As a result, vendors should thoroughly understand the system they are building and be able to respond to unexpected situations. Testing, performance limits, error messaging, reproducibility, and a strong maintenance plan are more critical than ever. OMB’s guidance specifically recommends demonstrations and testing that closely reflect the intended operating environment. Its guidance also addresses ongoing monitoring, independent evaluation, and the ability to return to a previous version when an update fails to meet established standards.

Support Transparency And Human Oversight

Depending on the application, government users will likely need to understand how AI-based results are produced, where limitations exist, and when human review is required. Even if a customer does not request these details, the vendor must certainly understand them in order to be able to effectively administer the system.

NIST’s AI Resource Center identifies several architecture and design mechanisms as important characteristics of trustworthy AI. It also emphasizes that trustworthiness depends on developers, subject matter experts, and oversight, not only the model.

Build Cybersecurity And Compliance Into The Design

In the realm of technology, widespread AI use is still fairly new, and the models themselves are incredibly complex. As a result, they can be unpredictable and represent a target-rich environment for adversaries, a vulnerability explicitly warned against in the international Guidelines for Secure AI System Development jointly published by CISA and the UK NCSC. Thus, cybersecurity should be designed into the systems being delivered to the government from the beginning.

Of course, security is also a contractual requirement in many situations. Depending on the environment, vendors may need to address ATO requirements, FedRAMP considerations, DFARS, NIST SP 800-171, and/or CMMC obligations where applicable.

Know The Difference Between Using AI And Engineering AI

It is easier than ever to build something that looks sophisticated.

A company can connect an application to a general purpose model, generate portions of the code, add a polished interface, and produce convincing responses quickly. That can be useful for prototyping, but it does not prove that the company understands how to build and operate a reliable AI system.

A capable vendor should understand model selection, data quality, retrieval methods, software architecture, evaluation, cybersecurity, failure modes, and human oversight. The vendor should be able to explain why technical choices were made and how they affect performance and risk, because they have deliberately evaluated those decisions rather than relying solely on AI-generated recommendations.

One of the benefits of AI that we’ve personally encountered is that it has opened up many new opportunities for small businesses like Kleos. Company size no longer determines capability to the level it did in the past. Smaller firms can bring specialized expertise, faster iteration, and closer customer collaboration. Still, the more important question is whether the people building the system understand it and can support their claims with evidence.

Plan For Portability And Transition

Federal acquisition guidance also addresses what happens after award. Specifically, OMB directs agencies to address data ownership, intellectual property, portability, knowledge transfer, licensing, and pricing transparency. It also encourages protections against arrangements that make changing providers prohibitively expensive. (OMB)

For example, who owns project data? Can the purchasing agency export its information? Can another contractor maintain the system? Is government data being used to improve a commercial product? What happens if the vendor changes its pricing or stops supporting the application?

Clear data formats, documented interfaces, reasonable ownership terms, and practical transition plans can make an AI system easier to sustain.

Conclusion

Federal AI acquisition appears likely to continue to create meaningful opportunities for vendors that can connect the technology to defined operational needs. Delivering technology for government missions is both an opportunity and a significant responsibility. After all, these systems may impact critical aspects of our daily life such as public safety, economics, infrastructure, and government decision-making.

At Kleos Technology, our approach begins with understanding the problem being solved. AI may provide part of the capability, but software architecture, data quality, custom logic, cybersecurity, testing, and human judgment are what turn that capability into a dependable product. We believe that these principles better position us to deliver products that are worthy of our customers’ trust.

Sources Cited

Back to all news