Skip to content
Vraelis/Docs
Getting started/AI security at Vraelis
Talk AI security
Overview
AI security at VraelisReview recorded task reportsGetting started with VraelisVerify, approve, run, re-checkWhat you can check
Connect an AI assistantThe command lineThe APIGate a release in CIWebhooks
Approving a planRun activityCompletionFindingsRepairRe-checks
SystemsGuaranteesMemory
← Back to vraelis.com
Implementation references

The product is in private development. Earlier browser-workflow examples describe previous implementation work. Console access, hosted APIs and execution services are closed.

Getting started

AI security at Vraelis

Our development scope, the security problems we are investigating and the foundations that exist today.

Outcome
Understand the current direction and distinguish research from implemented components.
Does not do
Public workspace access is closed. These references are not instructions for an available deployment.

Vraelis is developing independent cybersecurity software for AI-enabled defense, critical infrastructure and robotics. The company direction covers model integrity, adversarial threats, machine trust and security evidence.

Model integrity

Identify the model, dependencies and configuration a system is running, and compare them with reviewed artifacts. Authenticity helps establish provenance; it does not prove that a model is safe or free of backdoors. Runtime attestation and deployment integrations remain development work.

Adversarial threats

Investigate manipulated inputs, training-data poisoning and untrusted instructions reaching AI agents. Evaluation must name the model, attacker capabilities, conditions and limitations. We do not claim a universal attack detector or measured edge-inference performance.

Machine trust

Separate an AI proposal from permission to act. Identity, resource, action, workload integrity and policy must be established through trustworthy sources. Existing safety controls remain responsible for physical recovery and safe states.

Security evidence

Compare supplied reports against reviewed criteria and inspect the source events supporting a finding. Missing evidence remains unresolved. A recorded report is not authenticated telemetry or physical ground truth.

Current foundations

The repository includes a tested policy decision core and a recorded-evidence evaluator. These components are foundations for private integration work, rather than a complete operational security product. Trusted identity, artifact verification, adversarial evaluation and live enforcement require further development.

The private workspace address is data.vraelis.com. Sign-in and new account access are currently closed.

Earlier implementation referencesThe browser verification, CLI, API and console guides describe earlier implementation work. They do not define the current company scope or establish availability of those services.

Related

  • Review recorded task reports →
Next →Review recorded task reports
© 2026 Vraelisvraelis.comSecurityPrivacyCookiesTermsAcceptable use

On this page

Model integrityAdversarial threatsMachine trustSecurity evidenceCurrent foundations
CopiedCopy failed