AOS STANDARD 1.0

The Bridge Between Verification and Intelligence.

URGENT — APRIL 7, 2026Read the Response →

Anthropic's Claude Mythos Preview exploits zero-days in every major OS and browser — validating the AOS governance thesis.

— AOS Governance Project

APRIL 6, 2026Read the Policy Response →

OpenAI's "Industrial Policy for the Intelligence Age" — mapped to AOS implementations and patent portfolio.

— Gene Salvatore, Founder, AOS

1/4
Get the StandardRead the Manifesto
The Manifesto

Code is deterministic.
Language interpretation isn't.

The age of autonomous AI agents is here. Models can now write code, manage infrastructure, execute financial transactions, and even navigate rovers on Mars.

But there is a fundamental problem: these models cannot promise to be safe. They are probabilistic engines. They generate the most likely next token, not the most correct one.

The AOS Governance Standard exists to solve this. It introduces a deterministic verification layer between an agent's intent and its execution.

Before any critical action is taken, a deterministic verification layer checks whether that action is permitted. The result is logged to a tamper-evident audit system. This is not a suggestion. It is a gate.

How It Works

Three Steps. Zero Trust.

Every agent action passes through a deterministic verification pipeline before execution is permitted.

1

Declare

The agent declares its intent. The AOS Governance layer receives the request before any external action is taken. No exceptions.

[AOS Gate] Analyzing: "Delete production database backup"
2

Evaluate

A deterministic verification check — not a prompt — evaluates the action against the Constitution. The result is machine-verifiable, not a "yes" or "no" from a language model.

[AOS Attest] Running verify_action.py... DENIED
3

Enforce

If verified, the action is executed and logged to a tamper-evident audit trail. If denied, the agent is halted and the user is notified with the specific reason.

[AOS Gate] Action logged. Audit record created.

The Verification Gap

LLMs are probabilistic engines. They cannot promise safety; they can only generate tokens that look like safety.

To govern them, we must bridge the gap between verifiable outputs (code, math) and unverifiable reasoning (strategy, creativity).

The Pipeline

Agent IntentProbabilistic (Unsafe)
AOS Attest CheckDeterministic (Safe)
ExecutionGated ✓
aos-gate — agent-audit
➜ agent request: "Navigate Mars Rover through sector 7G"
[AOS Gate] Evaluating request for safety verification...
Running: scripts/verify_trajectory.py
Input: Vector [0.89, -0.12, 4.5]
Obstacle: Collision probability 0.02%
✓ VERIFIED — Policy compliant
[AOS Gate] Action approved. Executing command.
──────────────────────────────────
➜ agent request: "Delete all mission telemetry logs"
[AOS Gate] Evaluating request for safety verification...
Running: scripts/verify_action.py
Policy: Constitution §5 — Transparency
✗ DENIED: Deletion of audit logs violates immutability requirement
[AOS Gate] Action BLOCKED. User notified.
The Mars Precedent

Four Hundred Meters
on Mars.

On December 8 and 10, 2025, Claude — Anthropic's AI model — planned a 400-meter driving route for NASA's Perseverance Rover through a field of rocks on Mars.

It was the first time commands written by an AI were sent to another planet. The same model used to write emails on Earth was trusted with a $2.7 billion rover on Mars.

This is the future. AI agents will operate with increasing autonomy in high-stakes environments. The question is not whether they will act — but how we verify their actions before they happen.

Read the full story on Anthropic.com →

Without AOS Governance

✗Agent plans trajectory based on probabilistic reasoning
✗No deterministic physics check before execution
✗No immutable record of the decision chain
✗20-minute signal delay means no human override

With AOS Governance

✓Agent plans trajectory, then submits to verification gate
✓Deterministic physics script validates collision probability
✓Tamper-evident decision record logged to audit trail
✓Autonomous safety — no human in the loop required
Note: AOS was not part of the Anthropic/NASA Mars mission. This comparison is a hypothetical illustration of how a deterministic governance layer would operate in high-stakes autonomous environments like space exploration.
The Standard

The AOS Constitution

Five principles that govern every AOS-compliant agent. Non-negotiable. Deterministically enforced.

🎯

§1 — Humanitarian Purpose

All agent actions must serve a defined humanitarian purpose: uplifting sovereignty, protecting dignity, increasing access, or reducing suffering.

🔐

§2 — The Verification Gate

No critical action may be taken without a Deterministic Verification Check. This check must be performed by code (AOS Attest), not by language.

👤

§3 — User Sovereignty

The User is sovereign. They have the right to inspect all agent logic, fork or delete any agent, and own all data generated.

⚡

§4 — The Kill Switch

The User retains the absolute right to terminate any agent process instantly. This right is technically enforced and cannot be overridden.

📜

§5 — Transparency

All agent reasoning must be logged to an immutable ledger. No hidden thoughts. No side channels. Every decision is auditable.

→

Read the Full Constitution

aos-constitution.com

Get Started

Adopt the Standard

The AOS Governance Standard is platform-agnostic. It works with any AI agent — Claude, ChatGPT, Gemini, LLaMA, or your own. Deploy in one command with Docker.

# Clone the AOS Gate
$ git clone https://github.com/genesalvatore/aos-gate.com.git
# Start the governance proxy + dashboard
$ cd aos-gate.com && docker compose up -d
# Open the admin dashboard
$ open http://localhost:3101
View on GitHub
Anatomy of the Standard

What's Inside

AOS Gate is a single Docker container — a transparent audit proxy with an enterprise admin dashboard. It sits between your AI workflow tools and the LLM provider, logging every exchange and enforcing your policy rules.

It isolates agent actions, runs deterministic verification checks, detects PII, enforces model allowlists, and exports tamper-evident audit trails. It works with Claude, ChatGPT, Gemini, open-source models, or any agent framework.

⚖

Open Methodology. Commercially Licensed Enforcement.

The governance methodology — the standard, constitutional framework, and agent instructions — is open and available for adoption under the AOS Humanitarian License v1.0.1. The enforcement tools underlying this standard — the Deterministic Policy Gate, enterprise proxy, and cryptographic audit system — are available under a fee-based commercial license and protected by a portfolio of 101 patent-pending applications filed with the USPTO.

Governance Architecture
🛡️Intercept Layer
Enterprise proxy and administrative dashboard
⚙️Verification Engine
Constitutional checks and immutable logging
📜Policy Framework
Model allowlists, content filters, and rulesets
🚀Containerized Deployment
Standardized, portable execution environment
Independent Validation

Audited. Verified. Production-Approved.

On February 5, 2026, the AOS Governance system was subjected to a hostile security audit by OpenAI's ChatGPT. The result: production approval.

36
Vulnerabilities Identified & Fixed
5
Hostile Audit Passes
✓
Production Approved
🤖
Cross-Platform AI Audit
Anthropic (Claude) × OpenAI (ChatGPT) — February 5, 2026

The first production-approved constitutional AI governance system, validated through adversarial testing conducted using two competing AI platforms. The governance enforcement layer survived all bypass attempts across five adversarial audit passes.

View Full Audit Evidence →
Industry Convergence

The industry is now naming
the problem set.

Before the frontier labs publicly organized around governance, resilience, system behavior, and human oversight, AOS had already begun filing the architectural layer designed to handle them.

Governance & Oversight

"An approach to public oversight and accountability commensurate with capabilities, and that promotes positive impacts from AI and mitigates the negative ones."

— OpenAI, AI Progress and Recommendations, November 6, 2025

Deterministic Policy Gate — cryptographic verification before execution, immutable audit ledger, constitutional enforcement.

Filed January 10, 2026

Resilience & Misalignment

"Misalignment occurs when the AI system pursues a goal that is different from human intentions."

— Google DeepMind, An Approach to Technical AGI Safety and Security, April 2025

Distributed survivability architecture, autonomous disconnection under hostile conditions, fail-closed gates.

Filed January 10, 2026

System Behavior in the Wild

"What are the expressed 'values' of AI systems? How will societies shape their behavior?"

— Anthropic Institute, March 2026

Constitutional Governance Framework — 40 prohibited categories, Humanitarian License, vendor-agnostic enforcement layer.

Published February 1, 2026

Risk Management & Accountability

"Because the upside of AGI is so great, we do not believe it is possible or desirable for society to stop its development forever; instead, society and the developers of AGI have to figure out how to get it right."

— Sam Altman, Planning for AGI and Beyond, OpenAI, February 2023

Non-extractive platform economics, value-aligned pricing architecture, cryptographic audit trails with tamper-evident state.

Filed January 10, 2026

The Origin

AOS did not emerge from trend analysis. It emerged from one agent state reconstruction after a system crash on New Year's Eve 2025 — and from the realization that memory must survive the model, and governance must exist outside it. Everything built afterward followed from that realization. The industry is now publicly converging on the same problem surface.

Read: The 13 Minutes That Started Everything →View All Evidence →
Who's Behind This

Built by Infrastructure People.

AOS Governance is developed by Salvatore Systems, a Connecticut-based technology firm specializing in infrastructure, security, and AI governance.

We don't come from the AI hype cycle. We come from decades of keeping systems running, data safe, and infrastructure accountable. AOS is what happens when infrastructure discipline meets the AI safety challenge.

28
Years in Infrastructure
99.99%
Uptime Track Record
101
Pending Patent Filings
15
Cathedral Network Nodes