Skip to content
@evidencebound

EvidenceBound

Human control plane for verifiable AI agents: evidence, provenance, executable policy, deterministic verification, bounded blast radius and recovery.

EvidenceBound

Human Control Plane for Verifiable AI Agents

EvidenceBound is early-stage open-source AI safety infrastructure for binding agent actions to evidence, provenance and executable policy, with deterministic verification, bounded blast radius and recoverable human control.

The core design principle is simple: an AI system should not be able to grant itself trust. EvidenceBound keeps verification, policy enforcement and human authority outside model-generated claims and makes unavailable, stale or mismatched evidence visible rather than silently promoting it to certainty.

Core public implementations

Repository Purpose
evidencebound-core Framework-agnostic runtime for evidence/provenance binding, deterministic verification, policy enforcement and fail-closed selective recovery.
evidencebound-recovery-mesh Trust-aware selective recovery for autonomous agent fleets after a trust break.
evidencebound-authority-cut Reversible autonomy with human authority kept outside the model-callable surface and correction propagated through affected descendants.
evidencebound-releaseproof-dws Differential reverification for changed document evidence under a frozen historical review-equivalence policy.

What EvidenceBound is designed to preserve

  • Evidence visibility - missing, stale, blocked or mismatched evidence remains explicit.
  • Provenance - claims are bound to inspectable inputs and transformations.
  • Executable policy - trust decisions are enforced outside natural-language model output.
  • Deterministic verification - machine-checkable gates decide whether evidence and state remain acceptable.
  • Bounded blast radius - trust breaks invalidate only work that depends on the broken evidence or authority.
  • Recoverable human control - human grants, revocations and corrections remain outside model self-authorization.

Public identity

EvidenceBound does not claim that cryptographically authenticated provenance makes source data true, that an LLM can certify its own behavior, or that a working prototype is equivalent to a security-reviewed production system.

Pinned Loading

  1. evidencebound-recovery-mesh evidencebound-recovery-mesh Public

    Trust-aware recovery for autonomous agent fleets: detect trust breaks, compute dependency blast radius, reuse verified work, and selectively recompute affected branches on Google Cloud.

    Python

  2. evidencebound-core evidencebound-core Public

    Framework-agnostic runtime for evidence-bound AI agents: provenance, executable policy, deterministic verification, blast-radius analysis and fail-closed selective recovery.

    Python

  3. evidencebound-releaseproof-dws evidencebound-releaseproof-dws Public

    Differential reverification for document workflows: re-ground changed evidence and preserve prior human authority only under the review frozen equivalence policy.

    Python

  4. evidencebound-authority-cut evidencebound-authority-cut Public

    Reversible autonomy for professional AI agents: compute minimal policy-valid human authority and propagate later correction through reversible descendants without erasing unrelated safe work.

    Python

Repositories

Showing 5 of 5 repositories

People

This organization has no public members. You must be a member to see who’s a part of this organization.

Top languages

Loading…

Most used topics

Loading…