xMark: Forensic Provenance Watermarking

Executive Summary

xMark embeds forensic watermark payloads across multiple content dimensions using XorIDA threshold splitting. The watermark is not a single fragile signal — it is a distributed k-of-n secret that survives transcoding, compression, and re-encoding because no single transformation can destroy all shares simultaneously.

Two functions cover core watermarking: embedMark() splits a payload into threshold shares and embeds them across spatial, temporal, spectral, and metadata dimensions. Each share is independently robust against its expected transformations. verifyMark() extracts shares from all dimensions, validates HMAC integrity, reconstructs the payload from k surviving shares, and proves provenance.

Traditional watermarking hides data in statistical properties of media — when re-encoded, those properties change and the mark degrades. xMark distributes shares across independent dimensions where no adversary can attack all channels without destroying the content itself.

The implementation achieves Gold Standard Bronze — 922 lines of production code, 144 tests across 4 test files, per-share HMAC integrity verification, and xProve audit trail integration. Watermark reconstruction is mathematically guaranteed to succeed with k or more surviving shares, and provably fails with fewer than k.

The Problem

Traditional forensic watermarks embed identification data by manipulating pixel values, audio samples, or metadata fields. These marks can be detected, stripped, or destroyed through transcoding, re-encoding, or adversarial attacks. When the watermark is gone, provenance is lost.

Current watermarking systems face a fundamental trade-off: visible marks degrade viewing quality, while invisible marks are fragile against processing. A determined adversary can apply collusion attacks — comparing multiple watermarked copies to identify and remove the mark. Re-encoding degrades statistical properties where watermarks hide, causing mark loss before intentional removal even begins.

The Old Way

The PRIVATE.ME Solution

xMark splits watermark payloads into XorIDA threshold shares embedded across independent content dimensions — spatial (pixel), temporal (frame), spectral (frequency), and metadata. Each dimension carries a share that is independently robust against its expected transformations.

Verification reconstructs the watermark from any k-of-n surviving shares. This transforms forensic watermarking from a signal-processing problem into an information-theoretic one. Instead of hiding a mark that might be found and removed, xMark distributes shares across dimensions where no adversary can simultaneously attack all channels without destroying the content itself.

The New Way

How It Works

xMark combines XorIDA threshold splitting with multi-dimensional embedding, per-share HMAC integrity, and threshold verification for forensic provenance tracking.

Implementation Quality

Test coverage: 144 tests across 4 test files covering embedding, extraction, threshold reconstruction, integrity verification, and failure modes. All tests verify k-of-n threshold survival — watermark reconstruction succeeds when k or more shares survive, and correctly fails when fewer than k shares are available.

xProve integration: Every embedding and verification operation generates tier-appropriate verification proofs. Tier 1 HMAC chains provide tamper-evident integrity, while higher tiers (Beaver, IT-MAC, KKW) enable zero-knowledge verification without exposing the embedding algorithm or watermark payload.

Use Cases

Integration

embed and verify watermarks in two simple operations. HMAC verification happens before reconstruction — tampered shares are rejected before they can pollute the result.

import { embedMark, verifyMark } from '@private.me/forensicmark';

// Embed forensic watermark across 3 dimensions
const marked = await embedMark({
  content: videoBuffer,
  payload: { recipientId: 'critic-42', timestamp: Date.now() },
  threshold: { k: 2, n: 3 },
  dimensions: ['spatial', 'temporal', 'spectral']
});

// Verify provenance after transcoding
const result = await verifyMark(transcodedBuffer);
// result.value.recipientId === 'critic-42'

Security Properties

xMark guarantees information-theoretic watermark protection and tamper-evident integrity at every step.

Threshold Survival Guarantee

The k-of-n threshold property ensures watermark survival even when some shares are destroyed by transcoding or adversarial attacks. For example, a 2-of-3 watermark survives if any two of the spatial, temporal, or spectral shares remain intact. The implementation enforces this mathematically — reconstruction succeeds with k or more shares, and provably fails with fewer than k shares.

HMAC Integrity

Each embedded share includes an HMAC-SHA256 tag computed over the share data and metadata. During verification, HMAC validation happens BEFORE reconstruction — tampered shares are rejected before they can pollute the reconstruction process. This ensures that watermark extraction either succeeds with authentic shares or fails cleanly with clear tamper evidence.

Verifiable Data Protection

Every operation in this ACI produces a verifiable audit trail via xProve. HMAC-chained integrity proofs let auditors confirm that watermarks were embedded and verified correctly — without accessing the media or algorithm.

Honest Limitations

xMark is powerful within its threat model, but not suitable for all applications.

• Computational vs. information-theoretic: If an attacker has access to source code and implementation details, they can compute shares mathematically and mount collusion attacks. xMark assumes shares are embedded in independent dimensions that attackers cannot simultaneously access. In environments where an attacker controls the playback system, information-theoretic security degrades.
• Quality degradation at scale: Embedding multiple shares across spatial and temporal domains can introduce visible artifacts in demanding applications (scientific imaging, medical video). Test threshold parameters carefully for your content type. Audio watermarking is generally more robust than video across these dimensions.
• Steganography, not encryption: xMark proves provenance — it does not encrypt content. The watermark payload is encoded in shares, not encrypted. For confidential watermark data, combine xMark with encryption before embedding.
• Reconstruction requires k shares: If fewer than k shares survive a sequence of transformations, reconstruction fails. This is by design — it prevents attackers from reconstructing watermarks from partial information. Choose threshold parameters (k, n) based on expected survival rate through your distribution channel.

Ready to deploy xMark?

Talk to Sol, our AI sales engineer, or book a live demo with our team.