Security and trust

A clear security model for telemetry, analysis, and human review

Artemes AI is an early access security platform. This page documents the controls and boundaries implemented in the current architecture without claiming certifications, guarantees, or deployment properties that have not been independently verified.

Application records separated by customer and authenticated private routes

Deterministic analytics and sourced vulnerability reference data before AI review

Draft analysis, quality checks, human review, and separate promotion

Current data flow

Separate raw telemetry, deterministic analytics, and workflow state

The platform uses different stores for different responsibilities rather than treating model output as the source of truth.

Telemetry and replay

Fleet/osquery results flow through Pub/Sub into BigQuery, where raw observations support scheduled deterministic analytics and replay.

Sourced reference data

CISA KEV and FIRST EPSS data are ingested into reference tables and joined through reviewed rules that link software to CVEs.

Customer workflow state

Firestore stores assets, findings, context reviews, assignments, schedules, and dashboard summaries within each customer context.

Application access

Keep private product data behind authenticated routes

Marketing pages are public; customer and administrative workflows use a separate authenticated application.

Explicit customer context

Private reads and writes carry a customer identifier, and application records use Firestore paths for that customer.

Authenticated mutations

Administrative requests and API requests that change workflow state require authenticated credentials and authorization checks on the server.

Separate public forms

Early access and newsletter submissions use dedicated endpoints and collections rather than application data paths.

AI control boundary

Treat model output as a draft that must earn promotion

The analysis workflow is designed to retain provenance and block unsupported claims from silently becoming canonical findings.

Bounded input

The analysis step receives an asset context bundle and sourced facts; missing exposure, threat, outcome, or compliance evidence should remain missing.

Quality checks

Deterministic checks flag unsupported claims, and blocking failures prevent controlled promotion.

Practitioner review

Review state and promotion state are separate so approval and canonical updates remain explicit operational decisions.

Transparency

What this page does not claim

Security buyers should be able to distinguish implemented controls from future commitments.

No certification claim

Artemes AI does not currently publish a SOC 2, ISO 27001, FedRAMP, or other certification claim.

No absolute guarantee

AI output can be incomplete or incorrect and remains subject to practitioner review.

Responsible disclosure

Report a suspected security issue to info@gsconsultingllc.com with enough detail for the team to reproduce and assess it. Do not include unnecessary sensitive data.

Have a security or responsible disclosure question?

Contact GS Consulting directly. Please avoid sending credentials, production telemetry, or other unnecessary sensitive information by email.

Email the security team