Careers

AI and Machine Learning Engineer Jobs in Defense and National Security

Build the models and reasoning that turn fused mission data into understanding the customer can trust and trace.

AI and machine learning engineer jobs in defense build, evaluate, and field models that turn mission data into decisions people can trace, usually for employers whose work requires a U.S. security clearance or the ability to obtain one.

By the numbers. BLS projects U.S. employment of data scientists to grow 35 percent from 2025 to 2035, and of computer and information research scientists to grow 22 percent, citing demand for artificial intelligence (BLS Occupational Outlook Handbook). The median annual wage for computer and information research scientists was $140,300 in May 2025 (BLS).

This page is for engineers deciding whether to point their machine learning career at defense and national security. It covers what the work is, how it differs from commercial ML, what the jobs pay nationally and in Kansas City, how clearances work, and what the work looks like at Torch.AI.

What does an AI/ML engineer do in defense and national security?

The Department of War (DoW, formerly the Department of Defense) has its own definition. Its Cyber Workforce Framework includes a Data/AI workforce element, and within it an AI/ML Specialist work role (code 623) that "designs, develops, and modifies AI applications, tools, and/or other solutions to enable successful accomplishment of mission objectives" (DoW Cyber Workforce Framework, DCWF Tool v5.3, Sept. 2026). The same element defines an AI Test and Evaluation Specialist role (672) that tests AI "to ensure they are developed to be and remain robust, resilient, responsible, secure, and trustworthy."

In practice, the job is less about training the largest model and more about making a model's output usable for a decision with consequences. At Torch.AI that means building:

  • Entity resolution across many sources. Deciding when records from separate, multi-source systems refer to the same real-world person, unit, vessel, or object, and linking them into one picture that keeps the provenance of every link.
  • Answers grounded in a knowledge graph. Retrieval over a graph of resolved entities and relationships, so a model's response traces to source records instead of ungrounded text.
  • Deterministic, auditable logic. Encoding doctrine, policy, and rules as explicit, machine-evaluable logic that returns the same traceable result every time, with no opaque model in the critical path.
  • Model readiness, measured. Evaluating baseline, trained, and tuned models against real mission data so the customer fields evidence, not assumptions.
How it works

From scattered records to a decision you can trace

RECORDS FROM MANY SEPARATE SYSTEMSA DECISION A PERSON CAN DEFENDLINKPull records frommany source systemsRESOLVEDecide whichrecords are thesame entityGROUNDAnswer from a graphof resolvedentitiesEVALUATEMeasure the modelon real missiondataDECIDEA person reviewsthe traced answerhuman decidesEvery answer keeps its sources, so the person deciding can see why.

In defense work the model is one step, not the product. Entity resolution, grounding, and evaluation are what make its output usable for a decision with consequences.

NIST treats valid and reliable as the base of trustworthy AI, which is why evaluation on the customer's real data sits before any answer reaches a person.

Loops · respects reduced-motion · illustrative

How is defense AI/ML engineering different from commercial ML?

The algorithms are the same. The constraints are not. The comparison below is Torch.AI's view, anchored to the public standards that shape defense AI.

Dimension Typical commercial ML Defense and national security ML
Data Mostly first-party, consistent formats Fragmented across many systems, mixed formats and classification levels
What "good" means Engagement, conversion, accuracy on a benchmark A decision a person can defend, with the evidence behind it
Explainability Often optional Expected: the DoW's AI principles call for "transparent and auditable methodologies, data sources, and design procedure and documentation"
Human role Automation where possible Humans stay in command of consequential decisions
Where it runs Public cloud, always connected Classified enclaves and disconnected or degraded networks at the edge
Ownership Vendor-owned models and pipelines Increasingly government-owned systems the customer can inspect

Two public frameworks explain most of the right-hand column. The National Institute of Standards and Technology lists the characteristics of trustworthy AI as "valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed," and treats valid and reliable as "a necessary condition of trustworthiness" (NIST AI RMF 1.0, p. 12). The department's five AI Ethical Principles (2020) require AI to be responsible, equitable, traceable, reliable, and governable, including "the ability to disengage or deactivate deployed systems that demonstrate unintended behavior" (DoW release).

The department's 2023 Data, Analytics, and AI Adoption Strategy rests on an "AI Hierarchy of Needs" that starts with "quality data, governance, insightful analytics and metrics, assurance and responsible AI," and one of its stated goals is to "expand digital talent management" (DoW release, Nov. 2, 2023). Data quality comes first in that hierarchy, which is why so much defense ML work is data and evaluation work.

Build AI you can trace to source.

Entity resolution, GraphRAG, and deterministic logic on real mission data. See where your ML work fits.

Explainable models.
Real mission data.
Humans in command.

See open ML roles >

Entity resolution, GraphRAG, and deterministic logic on real mission data. See where your ML work fits.

Explainable models.
Real mission data.
Humans in command.

What do machine learning engineer jobs pay?

There is no federal occupation code for "machine learning engineer." The Bureau of Labor Statistics counts the work inside a few existing occupations, so the honest answer is a range across them. The table compares the national median with the Kansas City metro, where Torch.AI is headquartered.

Occupation (SOC) U.S. jobs U.S. median wage KC metro jobs KC median wage
Computer and information research scientists (15-1221) 37,200 $140,300 120 $104,670
Software developers (15-1252) 1,687,890 $135,980 12,160 $124,990
Data scientists (15-2051) 262,440 $120,230 1,260 $99,870
Computer occupations, all other (15-1299) 435,370 $116,580 3,900 $105,470

Source: BLS Occupational Employment and Wage Statistics, May 2025, national and Kansas City MO-KS MSA (28140). These are occupation-wide medians, not Torch.AI salaries.

What the numbers say:

  • Growth is strongest where AI work concentrates. BLS projects data scientists to grow 35 percent and computer and information research scientists 22 percent from 2025 to 2035, against 10 percent for software developers, and names AI as a driver for both research scientists and software developers.
  • Kansas City pays less in dollars and costs less to live in. Kansas City medians run below national medians, but Kansas City prices were about 7.5 percent below the U.S. average in 2024 according to BEA regional price parities. For more on the local market, see Kansas City AI jobs.
  • The catch-all category is more concentrated in Kansas City. "Computer occupations, all other" has a location quotient of 1.27 in the Kansas City metro, meaning it is 27 percent more concentrated there than nationally (BLS OEWS).

What skills do AI/ML engineers need for defense work?

The fundamentals are the same as anywhere: Python, statistics, model training and evaluation, and data engineering. Defense work raises the weight on a few others.

  • Evaluation and measurement. NIST puts validity and reliability at the base of trustworthy AI. Knowing how to measure a model on the customer's real data, and to say plainly where it fails, is a core skill rather than a final step.
  • Entity resolution and graph reasoning. Much mission data describes the same people, places, and things in different systems. Linking it correctly, with provenance, is often the hardest part of the problem.
  • Explainability by design. If an analyst cannot see why the system reached an answer, the answer is hard to use. Grounded retrieval and explicit rules help keep it traceable.
  • Working inside constraints. Disconnected networks, classified enclaves, and accreditation requirements shape how models are built, packaged, and updated.
  • Judgment about the human's role. The department's policy for autonomous and semi-autonomous weapon systems requires designs that "allow commanders and operators to exercise appropriate levels of human judgment over the use of force" (DoW release on DoD Directive 3000.09, Jan. 2023). Engineers who design for the person in the loop do better work in this field.

Do AI/ML engineer jobs in defense require a security clearance?

Many do, and the rules are set by federal regulation rather than the employer. Under 32 CFR 117.10, the government (the cognizant security agency) determines eligibility, a contractor may only request it when classified access "is essential" to a contract, and requests "will not be used to establish a cache of cleared employees." A contractor can start the process before your first day once it has made a written offer and you have accepted it in writing (117.10(f)).

That is why job listings say "active clearance required" for some roles and "clearable" for others. At Torch.AI, many roles require an active U.S. security clearance; others are open to clearable U.S. persons, with sponsorship for the right engineer. The full step-by-step path is in how to get a cleared AI job without a clearance.

What is it like to build AI at Torch.AI?

Torch.AI builds reasoning infrastructure for defense and national security: software that fuses data from many separate systems into conclusions a mission team can trace back to sources, in systems the customer owns. AI/ML here is not demos. It is reasoning that has to be right, explainable, and fast enough to matter when a decision is on the line.

Torch.AI was founded in Kansas City and is headquartered in Leawood, Kansas, with engineering roles there and in the Washington, DC region (see current openings).

Is this you? You will likely do well here if you:

  • Care as much about why a model is right as whether it scores well.
  • Want your work grounded in a mission, with humans in command of the loop.
  • Like building reasoning that survives contact with messy, real-world, adversarial data.
  • Hold or can hold a U.S. security clearance.

This may not be for you if you want to chase benchmark numbers in a sandbox, you are indifferent to explainability and provenance, or you would rather stay far from how the model gets used.

Related paths: AI Software Engineer, AI Data Engineer, and DevOps and Platform Engineer.

How do you apply for AI/ML engineer jobs at Torch.AI?

Current openings, locations, and clearance requirements are listed at torch.ai/careers. Each listing states whether the role requires an active clearance or is open to clearable candidates. Tell us what kind of reasoning you want to build and why it matters to you.

Put your models on the mission.

We hire ML engineers who care why a model is right, not just how it scores. Find your opening.

Cleared or clearable.
Government-owned AI.
Work that ships.

View ML openings >

We hire ML engineers who care why a model is right, not just how it scores. Find your opening.

Cleared or clearable.
Government-owned AI.
Work that ships.

Sources

See open roles