WitnessBound — Authority Lineage for AI Governance
Companion to IBA

WitnessBound
Witness-Bound Authority — WBA

IBA proves the agent acted within scope.
WBA proves the scope was legitimately declared.

When an AI agent acts on signed intent — who authorized that signature? Was it freely given, by the right principal, under non-coercive conditions? Can you prove it to a regulator eighteen months later?

The Unsolved Problem
01 · Cryptographic validity is not legitimacy.

A re-signing event that mutates an AI agent’s authorized scope can be cryptographically perfect — correct key, valid signature — while the authority behind it was coerced, compromised, or obtained under operational pressure. The signature is genuine. The authorization was not.

02 · The enforcement layer needs a foundation.

IBA ensures agents act within declared intent. But if the declaration itself was illegitimate, IBA faithfully enforces the wrong boundary. Runtime enforcement cannot fix a corrupted authority chain.

03 · Boards will ask the lineage question.

When something goes wrong with an AI agent in a regulated environment: “who authorized that scope, when, under what conditions, and can you prove it?” No current framework has a complete answer.

04 · Evidentiary survivability is a design requirement.

The authority lineage of an AI agent deployment must be reconstructible — without ambiguity — months or years after the fact. As a first-class architectural requirement, not a logging afterthought.

The WBA Function

Every signing event is itself a governance object

WBA(SigningEvent, Principal, Context, Time) = f(
  AuthorityProof,   // Is this principal authorized to sign?
  ContextualAnomaly,   // Does this event deviate from baseline?
  WitnessThreshold,   // How many co-signatories required?
  EvidentiaryChain   // Is the lineage reconstructible?
)
01 · ABE

Authority Baseline Engine

Maintains a continuously updated behavioural profile for each principal authorized to sign intent declarations — signing times, contexts, scope delta magnitudes, frequency, co-signatory patterns.

02 · CAD

Contextual Anomaly Detector

At each signing event, computes an Anomaly Score against the principal’s baseline profile. High score triggers elevated Witness Threshold.

03 · WTE

Witness Threshold Engine

Configurable by deployment context. Low-stakes re-signings proceed with single-principal authority; high-stakes, high-anomaly events require multi-principal co-signing or temporal delay.

04 · SCR

Signing Context Record

Every signing event produces an immutable record — principal identity, timestamp, anomaly score, witness threshold applied, co-signatories, intent hash. Cryptographically chained to all prior records.

05 · ERA

Evidentiary Reconstruction API

Produces a complete, verifiable authority lineage for any deployment, at any future point. Formatted for regulatory submission and legal proceedings.

The Complete Governance Stack
LayerQuestion AnsweredIP Status
IBADid the agent act within declared scope at runtime?Patent Pending GB2603013.0
WBAWas the scope declaration itself legitimately authorized?Filing proposed Q1 2026
IP Status
AssetDetail
IBAPatent Pending GB2603013.0. Filed 10 February 2026, UK IPO.
WBAFiling proposed Q1 2026. Authority lineage verification — contextual anomaly detection, witness thresholds, Signing Context Records, evidentiary reconstruction.
NISTNIST-2025-0035. Filed 19 February 2026. On the public record — a filing, not an endorsement.

The patent is filed. The companion is being drafted. Acquisition enquiries, technical collaboration, and licensing discussions welcome.

Contact

IntentBound.com · Patent GB2603013.0 (Pending) · UK IPO · PCT 150+ Countries