AI security · Authority boundaries · Adversarial validation

AI systems need enforceable boundaries.

Independent security engineering for AI agents: authority mediation, adversarial testing, state integrity and evidence-based delivery.

Explore services Review public evidence

01 / Portfolio

Solutions

Research and engineering capabilities, with commercialization limited to qualified scope.

Surgical DevOps

Governed execution and authority boundaries for AI-assisted development.

Engineering / pilot services

Custos

Experimental, non-sovereign authority mediation and validation layer.

Qualification in progress

Enterprise Agent Security Lab

Bounded adversarial experiments on permissions, tools, persistence and recovery.

Supervised assessments

SACP / F´ research

External command-mediation experiment around unmodified F´ v4.4.1.

Research — not flight qualified

02 / Research

Public evidence & external technical engagement

Evidence is version-specific. External engagement is not institutional certification or endorsement.

AGMI Pull Request #8 ↗

Surgical DevOps G7 integrity contribution merged through AGMI PR #8 on October 15, 2026. Independent macOS testing reported detection of nine bounded tampering scenarios on the read path, within the stated attacker scope.

Independent Architectural Review — NASA cFS ↗

Independent collaboration with Rahad Bhuiya on Community Architecture Proposal v0.1 for the NASA cFS ecosystem. Thales Rangel co-authored the architectural foundation, performed independent documentary review, analyzed responsibility boundaries, and preserved traceability across 33 requirements and six security invariants. The documentary baseline was merged through PR #2; all six invariants remain DRAFT / NOT_QUALIFIED. Independent research with no institutional affiliation, NASA certification, or endorsement.

Community Architecture Proposal v0.1 ↗ · cFS discussion #1168 ↗

Surgical DevOps repository ↗

Public source, review entry points and engineering history; examine exact commit and limitations.

Pinned qualification run ↗

Historical, commit-specific CI run referenced in the review materials; not a universal platform certification.

03 / Commercial

Enterprise services

Initial engagements are fixed-scope, authorized and evidence-driven.

AI Agent Security Boundary Assessment

Review permissions, tool invocations and trust boundaries in a permitted test environment.

Scoped service

State Integrity & Recovery Review

Evaluate bounded tamper/replay/restart scenarios with reproducible evidence.

Scoped service

04 / Contact

Request a scoped assessment

Contact through the project GitHub profile while a dedicated business channel is being established.

Open Bonushora GitHub ↗

No security guarantee, flight-readiness claim, NASA/JPL or OWASP certification is implied. No public cloud execution environment is enabled at this stage.