More than 4,200 federal AI roles were posted on USAJobs.gov for new-grad-eligible candidates across Q1 and Q2 2026 — a figure that does not appear in any university placement report ENTRA reviewed, and that most CS career offices are not actively surfacing to graduating seniors. The agencies doing the hiring include DARPA's Information Innovation Office, the Department of Energy's national lab network, the Center for AI Standards and Innovation at NIST, and the Army Futures Command's AI Integration Center. The base pay runs $95,000 to $145,000 — below frontier lab offers — but the clearance premium, mission specificity, and structural job security are pulling a measurable share of the Class of 2026 away from the San Francisco pipeline.
Government AI Hiring 2026: Who Is Actually Recruiting for Federal AI Roles
DARPA's Innovation Fellowship is the most selective federal entry point for early-career researchers. The program primarily targets recent PhD graduates (within five years of receiving a doctorate), with exceptions possible for bachelor's and master's-level candidates in exceptional cases. Fellows are appointed for one year — extendable to two years — based at Arlington, Virginia, embedded directly within DARPA's program offices. The Information Innovation Office, which houses DARPA's AI, cybersecurity, and autonomy programs, is the primary home for CS and ML candidates. Fellows work inside active DARPA programs, not support functions. Eligibility requires U.S. citizenship and the ability to obtain a security clearance; prior clearance is not required at application. DARPA does not publish a salary figure for fellows, but the GS equivalent for the role lands in the GS-11 to GS-13 band — $95,000 to $125,000 base in the Washington, D.C. locality, before the 33.94% D.C. locality adjustment that OPM applied in the 2026 pay tables.
The Center for AI Standards and Innovation (CAISI) — which Secretary of Commerce Howard Lutnick reconstituted from the original NIST AI Safety Institute in June 2025 — is now staffing aggressively across two locations: downtown San Francisco and Washington, D.C. CAISI has published seventeen specific taskings under the Trump administration's AI Action Plan, and the technical work it requires maps directly to skills the Class of 2026 developed through RLHF contracting and model evaluation work at Anthropic, Scale AI, and Outlier. The center is hiring software engineers, AI research engineers, AI research scientists, and specialists in measurement and validation of AI systems. Applicants can submit interest through a public Google Form; some positions are issued under direct-hire authority, which bypasses the standard competitive civil service process and compresses time-to-offer substantially. CAISI operates from a GS-12 to GS-14 pay band for technical research roles — $107,000 to $146,000 base in San Francisco with locality applied.
DOE's national lab network — Argonne (Lemont, Illinois), Oak Ridge (Oak Ridge, Tennessee), Sandia (Albuquerque and Livermore), Los Alamos (New Mexico), and Lawrence Berkeley (Berkeley, California) — is the largest single employer of new AI and ML researchers in the federal ecosystem. Argonne received fresh DOE funding in 2025 specifically to advance AI for science, including projects integrating AI with advanced robotics for protein design alongside Berkeley, Oak Ridge, and Pacific Northwest National Laboratory. OpenAI deployed models at Los Alamos in early 2025 as a shared research resource across Los Alamos, Lawrence Livermore, and Sandia — a partnership that accelerated internal ML tooling and created new demand for engineers who can operate at that interface. Sandia's median total compensation for research scientists sits at $105,000 to $131,000 per year, per Levels.fyi and Glassdoor data through April 2026. Software engineers in the Member of Technical Staff classification run $120,000 to $187,000 depending on level. The national labs do not compete with frontier lab equity packages — but they offer something frontier labs structurally cannot: indefinite, mission-driven employment with government-backed benefits, pension accrual, and physical research infrastructure that no private lab has replicated.
MITRE Corporation's New Professionals program represents a second class of federal-adjacent hiring: Federally Funded Research and Development Centers (FFRDCs) that operate under government contract but hire as private employers. MITRE explicitly targets candidates with fewer than two years of post-graduation experience. Entry-level data science and AI roles pay $80,000 to $120,000 annually. The work covers cybersecurity, AI safety, and defense systems analysis — the same problem surface that CAISI and DARPA are working, but with MITRE's managed overhead structure handling clearance processing, compliance, and infrastructure.
The Army Futures Command's AI Integration Center (AI2C) — headquartered at Carnegie Mellon University in Pittsburgh, Pennsylvania, under Army Futures Command, which is based in Austin, Texas — added a formal AI and machine learning officer career path in December 2025. Defense Secretary Pete Hegseth directed the Army to field AI-driven command and control at theater, corps, and division headquarters by 2027 — a mandate that is pulling software engineers and ML researchers into Army Futures Command roles at a rate the Army's traditional civilian hiring pipeline was not built to absorb. The AI2C is hiring civilian software engineers alongside the uniformed officer track; target start dates for civilian roles are running Q3 2026.
Federal AI Salary vs Frontier Labs: Why the Pay Gap Is Smaller Than It Looks
The headline gap between federal AI comp and frontier lab comp is real: $120,000 federal base versus $185,000 total compensation at OpenAI or Anthropic is a material difference for a 22-year-old carrying student debt. But the headline gap narrows significantly once clearance premium and contractor market access are factored in.
A Secret clearance — the minimum level most federal AI roles require, typically processed within 6 to 12 months of hire — adds $20,000 to $40,000 to annual compensation in the cleared contractor market, per ClearanceJobs 2026 data. A TS/SCI adds $20,000 to $50,000 beyond that. The calculation that matters for a Class of 2026 graduate is not the year-one number. It is the year-three number, when the clearance is active and portable.
Defense contractors are the mechanism through which that premium converts to cash. Booz Allen Hamilton posted a Mid AI/ML Engineer role in Herndon, Virginia in March 2026 at up to $176,000 annually. Its Senior AI/ML Engineer at Fort Meade, posted April 23, 2026, runs $99,000 to $225,000 depending on clearance level. Leidos is expanding its DOD support contracts and pulling ML engineers into cleared roles at rates competitive with mid-tier tech companies. SAIC's AI practice is running similar salary bands across its defense intelligence and cyber programs. These are not clearance-holder retention bonuses — they are the standard market rate for an engineer who can work in a classified environment.
In-Q-Tel adds a third pathway. The CIA's venture capital arm has made nine investments in 2026 alone through May, seeding AI startups — including portfolio companies in autonomous systems and AI compute — that then hire engineers directly. In-Q-Tel portfolio companies often process clearances as a standard onboarding step, creating a private-sector career path that sits inside the cleared ecosystem without requiring a direct government hire.
The total picture: a Class of 2026 graduate who takes a $110,000 CAISI research engineer role in San Francisco, processes a Secret clearance over 12 months, and moves to a Booz Allen or Leidos cleared role in year three is on a $165,000 to $200,000 trajectory — without the equity cliff risk that accompanies frontier lab offers.
Which 2026 Graduates Should Pursue DARPA, NIST, and DOE AI Careers
Federal AI hiring is not the right call for every CS graduate. The constraints are specific: you need U.S. citizenship, the ability to pass a background investigation, and a tolerance for the security restrictions — no public publishing without review, no simultaneous consulting for private AI labs, travel limitations — that come with cleared work. The pace of frontier lab research is faster; the infrastructure for training large models lives at OpenAI and Google, not at Sandia.
But for a subset of the Class of 2026 — particularly graduates from safety-focused programs, students who prioritized evals and interpretability research over benchmark racing, and candidates who are uncomfortable with frontier labs building systems toward AGI under commercial timelines — federal AI represents a coherent alternative that most career offices have not presented to them.
CAISI's evals mandate is the clearest case. The center's technical staff is charged with developing evaluation methodologies for AI systems deployed in sensitive contexts — work that is nearly identical to what RLHF evaluation contractors did at Anthropic and OpenAI, but with a public interest mandate and without the model-deployment pressure. Graduates who spent six to twelve months on Outlier or Mercor's coding eval tracks are holding the exact skill set CAISI needs. CAISI's direct-hire authority means that skill set can translate to an offer in weeks rather than the standard four-to-six month federal hiring cycle.
The DARPA Innovation Fellowship has a narrower profile requirement — it is looking for candidates willing to work inside the agency's program management structure, not just execute research — but its two-year tenure creates a federal career credential that opens every subsequent door in the defense-AI contractor market.
Three things to watch in the second half of 2026: whether CAISI's San Francisco office materially expands its headcount as it executes the AI Action Plan taskings; whether Army Futures Command's AI2C civilian track begins competing directly with Austin's tech employer base for ML talent; and whether the DOE national labs, which collectively represent the largest physical AI compute infrastructure outside of the hyperscalers, begin advertising their postdoc-to-staff conversion pipelines more aggressively to the same CS programs OpenAI and Anthropic are recruiting from.
The 4,200 open roles are not a rumor. They are posted, they are funded, and most of the Class of 2026 has not looked at them.
Sources: DARPA Innovation Fellowship — CAISI / NIST AI Safety Institute — OPM 2026 GS Pay Scale — Sandia National Labs Salaries, Levels.fyi — Glassdoor Sandia Salaries 2026 — Argonne AI Funding, ANL — Booz Allen AI/ML Engineer, Herndon VA — TS/SCI Salary Premium 2026, ClearanceJobs — In-Q-Tel Portfolio 2026, Tracxn — Army AI/ML Career Path, Federal News Network — MITRE Early Career Programs — USAJobs AI Roles — DOE Jobs at National Labs