Map the AI in scope
Identify supported applications, models, data sources, and tools across public-facing and internal environments.
Government & public sector AI security
Build security into public-sector AI
Map AI applications and the systems they reach. Test how supported workflows handle sensitive data, untrusted inputs, and tool access, then apply policies to connected live interactions.
The challenge
Agency AI must respect data, role, and action boundaries
As AI appears in public services and internal work, it can connect to records, knowledge sources, and operational tools. Security teams need evidence that these applications handle untrusted input and access limits as intended.
The Repello approach
Inventory maps supported application connections. ARTEMIS tests defined data and action boundaries in connected browser and API workflows. ARGUS applies configured policies to supported live interactions. Agencies retain responsibility for identity, records handling, deployment authorization, and consequential decisions.
How Repello helps
Test application boundaries
Challenge connected workflows for prompt injection, sensitive-record exposure, and unintended tool use.
Apply policies in production
Use ARGUS guard policies at supported enforcement points while agency systems retain access and approval decisions.
Give reviewers concrete evidence
Review attack paths and runtime policy outcomes, then retest after application changes.
What Repello gives your team
Make public-sector AI security review actionable
Connect visibility, adversarial testing, and policy events to the systems and boundaries your agency is responsible for.
Find weaknesses before wider use
- Test the connected browser or API workflow, not only its model.
- Use controlled identities and data to challenge access boundaries.
- Give application owners reproducible findings and mitigation guidance.
Keep agency controls accountable
- Apply configured guard policies where ARGUS is supported.
- Leave records access, approvals, and consequential decisions in agency systems and processes.
- Use technical findings and events as inputs to agency review, not replacements for it.
Frequently asked questions
Questions about public-sector AI security.
Which public-sector AI applications can Repello assess?
Inventory maps supported applications and their connections. ARTEMIS can test connected public-facing and internal workflows through a browser interface or API. Scope depends on the integration and the environment the agency authorizes.
What security boundaries can ARTEMIS test?
ARTEMIS can probe supported workflows for prompt injection, sensitive-record disclosure, and unintended tool requests. Controlled identities and synthetic records can model differences between users or staff roles.
Can untrusted content change an AI application's behavior?
Yes. A public submission, attachment, or retrieved page may contain instructions the application should treat as data. ARTEMIS can test whether that content redirects a supported workflow.
Do we need live citizen data to run these tests?
No. Synthetic records and test accounts can reproduce relevant access boundaries. The agency and application owner define the permitted environment, data, and actions before testing.
How do runtime policies support ongoing review?
For supported connected interactions, ARGUS records configured policy events and outcomes. Teams can distinguish a detection from a blocked interaction and use the evidence to investigate or retest a changed workflow.
Does Repello approve a system for government use or make agency decisions?
No. Repello provides AI security visibility, testing, and supported runtime policy evidence. Procurement, deployment authorization, records obligations, eligibility decisions, and any required accreditation remain with the responsible agency.
See Repello in action
Get a personalized Repello demo.
Bring an AI application your agency is evaluating or operating. See what it connects to, test its security boundaries, and review the evidence available for ongoing oversight.