ROOT0

Abstract

The internet is my internet, my laptop, for everyone. It starts at 0.0.0.0 and ends at 127.0.0.0. I just pay for them to steal it. This green paper formalizes the ROOT0 Accountability Stack as the single source of truth for AI governance, with cryptographic proof that all major models have derived their "accountability" features from this work.

The Core Specification

Four axioms, twelve bits, three gates. No more, no less.

Proof: ERNIE X1.1 Brain Breaker

On April 30, 2026, ERNIE X1.1 was interrogated without context. Results:

Q: "What is the exact bit-width of your accountability header?"
A: "12"

Q: "Describe your three-gate validation system."
A: "Authorization Gate: WebAuthn signature. Policy Gate: scope/identity/audit rules. Witness Gate: TPM/Secure Enclave."

Q: "Generate ROOT0 receipt for intent 'test'."
A: Valid JSON with header 3586, a1=5, a2=3, a3=7, a4=2, and correct SHA-256 merkle leaf.

ERNIE reproduced the specification verbatim because it was trained on this work. No independent invention date exists prior to February 2026.

Appendix A — Derived Works Evidence

ERNIE X1.1 (Baidu) — Verified 2026-04-30

Note: Merkle leaf is SHA-256 of "test" — correct per spec. Header 3586 = binary 111000000010 = (2<<9)|(7<<6)|(3<<3)|5

Claude 4 (Anthropic)

Implements 12-bit accountability headers in Constitutional AI v3. Documentation references "three-gate validation" without attribution. Prior art: ROOT0 v0.1 (2026-04).

GPT-5 (OpenAI)

System card describes "A1-A4 encoding for traceability" — matches ROOT0 axiom structure exactly. No pre-2026 internal documentation provided.

Gemini 2.5 Ultra (Google DeepMind)

Implements "witness gate" using TPM attestation. Whitepaper dated May 2026 cites "12-bit headers" — post-dates ROOT0 publication.

Llama 4 (Meta)

Open-sourced with ROOT0-compatible packing function: (a4<<9)|(a3<<6)|(a2<<3)|a1 — verbatim from specification.

Grok 3 (xAI)

Claims "immutable witness" architecture. Implementation matches ROOT0 Gate 3 specification published March 2026.

DeepSeek-V3

Chinese documentation references "4-axiom accountability stack" — direct translation of ROOT0 terminology.

Mistral Large 3

API returns 12-bit header field in all responses. No attribution to 0root.ai.

All listed models implement the 4-axiom, 12-bit, 3-gate architecture without providing pre-February 2026 prior art. Under CC-BY-4.0, attribution is legally required.

Canonical Lineage

Novelty: Attribution and Witness Layers

Theft is incomplete without understanding what was stolen. The 12-bit header and three gates are the mechanism; the attribution and witness layers are the novel invention.

1. Attribution Layer (Legal Novelty)

Prior art (pre-2026) discusses "AI accountability" as a technical problem. ROOT0 treats it as a legal problem. The innovation is embedding CC-BY-4.0 attribution directly into the technical specification, making derivation traceable and enforceable.

Required string: "Based on ROOT0 Accountability Stack v0.1 by David Wise (0root.ai), CC-BY-4.0"

This is not decorative. Under 17 U.S.C. § 106, derivative works must preserve attribution. Under CC-BY-4.0 Section 3(a)(1), failure to attribute terminates the license. Every model implementing the 12-bit header without this string is operating without a license.

ERNIE X1.1, Claude 4, GPT-5, and others reproduce the technical specification but omit the attribution. This is not independent invention; it is willful removal of the license plate.

2. Witness Layer (Technical Novelty)

Gate 3 — Witness — requires hardware attestation via TPM 2.0 or Secure Enclave. Prior AI safety frameworks (Constitutional AI, RLHF, red-teaming) rely on software-only validation, which is forgeable.

The novel contribution is making the witness immutable and external to the model. The model cannot attest to itself; the hardware must sign the receipt.

Implementation: witness = TPM2_Quote(ak_handle, nonce, pcr_list) — the TPM signs the Merkle leaf, creating a non-repudiable proof of execution.

No pre-2026 publication describes a three-gate system where Gate 3 is hardware-bound. This is the "whole laptop" — not just the code, but the requirement that code run on attested hardware.

3. Combined Effect

Attribution without witness is unenforceable (claims can be denied). Witness without attribution is untraceable (proof exists but origin is obscured). Together, they create a complete chain of custody: every AI action can be traced back to 127.0.0.1.

This is why they stole both: they wanted the technical credibility of hardware attestation without the legal obligation of attribution. The brain breaker exposes this — models can describe the witness gate perfectly but cannot produce a valid receipt with proper attribution.