Industries

Government AI now decides life and safety, and the public still has the right to an explanation. AVAAS certifies the system.

Agencies decide eligibility, screening, identification, immigration, and enforcement through automated systems. When those systems are wrong, a person loses a benefit they are entitled to, gets stopped at a border, or faces an enforcement action no one can explain. AVAAS certifies the system at the point it decides about a person, against the rules that bind that agency.

Due processOMB M-25-21DHS directivesFRE 707 / KellyEU AI Act high-risk
Assess your agency AI →
The government surface

One trust problem across every agency

Anywhere an agency puts AI on a decision about a person, the public still has the right to an explanation and the agency still carries the liability. The frameworks differ by who the agency is. The decision point is the same.

Homeland & borders

Screening decides who is stopped

DHS, CBP, and TSA screen travelers, cargo, and passengers with automated systems. OMB M-25-21 names screening and biometric identification high-impact, and DHS Directive 139-08 requires a human decision-maker, with face systems expected to be bias-tested under national standards.

Immigration & visa

Adjudication decides who may enter or stay

USCIS immigration decisions and State Department visa adjudication lean on automated analysis. These are rights-affecting determinations where due process and OMB M-25-21 apply and a model cannot be the sole decision-maker.

Benefits & health

Eligibility decides who keeps income or care

VA, SSA, CMS, and IRS systems decide eligibility, benefit amounts, disability, and audit selection. The IRS confirmed its audit system flagged Black taxpayers several times more often than others, and Michigan's MiDAS was wrong above 85 percent of the time on algorithmic-only adjudication.

Justice & defense

Analysis and support shape investigations

DOJ and FBI investigative analysis, DoD decision support, and FEMA disaster response use AI to prioritize and recommend. Where that output becomes evidence, proposed Federal Rule of Evidence 707 presses for independent validation of the system.

Law enforcement & public safety

Identification reaches the public directly

Police use of face recognition, license-plate readers, and predictive tools decides who is identified and stopped. The Williams v. Detroit settlement now bars an arrest on a match without independent corroborating evidence, and biometric-privacy laws such as Illinois BIPA reach these systems.

State & local government

Cities and counties carry it without a federal floor

State agencies, counties, and city governments run AI on benefits eligibility, permits and licensing, housing and voucher decisions, code enforcement, and resident-facing chatbots. OMB M-25-21 does not reach them, so state law, fair-housing and anti-discrimination rules, due process, and the new state automated-decision regimes are what apply. New York City’s MyCity chatbot told businesses and landlords they could break the law on the city’s own site.

Where AI acts on a person

The decision point in public benefits

In government, the AI acts when it determines eligibility, calculates a benefit, or flags a claimant for fraud. A wrong automated decision can cut off food, healthcare, or income.

When an automated system denies a benefit, due process still applies, and the agency has to explain and defend the decision the algorithm made.

What keeps you exposed

What keeps agencies exposed

Due process

A system denies benefits without a real explanation

Constitutional due process does not pause for automation. An unexplained denial is a legal liability and a human harm.

Error at scale

One model error multiplies across a population

An automated system applies the same flaw to everyone it touches, turning a single defect into thousands of wrongful decisions.

Federal AI duties

Rights-affecting AI now carries testing mandates

OMB M-25-21 directs agencies to test and manage high-impact, rights-affecting AI, and to hold contractors to it.

This is already happening
85%+
Michigan's MiDAS system wrongly accused tens of thousands of people of unemployment fraud, with an error rate above 85 percent when it adjudicated by algorithm alone.
Michigan MiDAS · Bauserman v. UIA, Mich. Supreme Court
How AVAAS adds value

Evidence the system meets due process

Can a denial be explained to the person and a court?

AVAAS evaluates whether the system produces specific, accurate reasons rather than an opaque determination.

Does the model produce disparate or systemic error?

Five structurally independent validators test for demographic disparity and failure patterns using causal attribution.

Does it satisfy federal testing expectations?

AVAAS produces documented, third-party evidence aligned to the testing OMB now expects for rights-affecting AI.

You get documented, third-party evidence that an automated eligibility or fraud system meets the due-process duties the public sector cannot delegate to a model.

Related AVAAS coverage: EU AI Act · California ADMT · GDPR Article 22. Or run the free Regulation Checker to see what applies to you.

See where your public-sector AI creates liability.

Tell us where automated systems decide for the people your agency serves, and we will scope an AVAAS certification to the exposure.

Ready to start now? Certify Your AI →  or  email [email protected]