Verifiable Surveillance Data: Privacy-Aware Architectures

Short description: Decision-focused playbook for building provable, privacy-preserving surveillance data pipelines using today’s standards: in-device authenticity, remote attestation, selective-disclosure credentials, transparent logs, and cost-efficient data availability layers.

Why this matters in 2025

Surveillance data (cameras, access control, IoT telemetry, vehicle sensors) now feeds high‑risk AI and compliance workflows. Buyers increasingly demand two things at once: cryptographic proof that data is authentic and unaltered, and strong privacy guarantees that limit exposure of people and places. The standards landscape finally supports this dual mandate:

W3C Verifiable Credentials 2.0 reached Recommendation on May 15, 2025, enabling interoperable, machine‑verifiable proofs across wallets, devices, and services. (w3.org)
EU AI Act codifies mandatory logging for high‑risk AI: systems must record events over their lifetime (Art. 12) and keep automatically generated logs for at least six months (Art. 19). (ai-act-law.eu)
California’s new CPPA rules finalize risk assessments, cybersecurity audits, and Automated Decisionmaking Technology (ADMT) obligations starting January 1, 2026—directly affecting video analytics and telemetry decisioning. (cppa.ca.gov)
Device/compute attestation is standardized: IETF’s RATS architecture (RFC 9334) and the Entity Attestation Token (EAT, RFC 9711) let verifiers cryptographically assess camera, gateway, or enclave state. (ietf.org)
Content provenance moved on‑device: C2PA 2.0/2.2 brought firm cryptographic baselines; Leica and Sony now ship capture‑time Content Credentials/C2PA in production firmware and offerings. (spec.c2pa.org)
Data availability got cheaper: Ethereum’s Dencun (EIP‑4844) added 18‑day “blobs” for inexpensive rollup data; modular DA layers like Celestia/Avail and operator DA like EigenDA broaden options. (ethereum.org)

Below is a concrete architecture template and implementation patterns we deploy for startups and enterprises evaluating blockchain-backed surveillance systems.

Design goals (and non‑goals)

Verifiable end‑to‑end provenance: cryptographic chain from lens/sensor to report.
Privacy by default: selective disclosure, minimization, and enclave‑guarded analytics.
Interoperability: VC 2.0 + OpenID4VCI issuance; DID‑addressable devices/organizations. (w3.org)
Transparent but non‑leaky auditability: public/verifiable logs without exposing raw PII.
Cost control at scale: DA choice (Ethereum blobs vs Celestia/Avail/EigenDA) by retention/SLA.

Non‑goals:

Full on‑chain storage of video/audio. Store hashes, manifests, and proofs—never raw streams.

A reference architecture for verifiable, privacy‑aware surveillance

Think in seven interoperable layers. Each layer swaps in standards‑based components.

Capture and attestation (device and compute)

Cameras/sensors sign media at capture using C2PA Content Credentials; when available, bundle device attestation (EAT). (c2pa.org)
If preprocessing or analytics run in a TEE (e.g., AWS Nitro Enclaves), collect attestation documents (CBOR/COSE) proving enclave identity and code hash (PCRs 0–4,8). (docs.aws.amazon.com)

Identity and authorization

Address devices and organizations with DIDs; issue access/processing entitlements as W3C VCs using OID4VCI. Prefer selective‑disclosure formats (BBS+ Data Integrity or SD‑JWT VC) for minimal data release. (w3.org)

Ingest and privacy‑preserving transport

Use Oblivious HTTP (RFC 9458) relays for IP‑unlinkable telemetry submission (client↔relay↔gateway split). (ietf.org)

Analytics with verifiability

Run watchlist matching or face‑blur pipelines inside TEEs; gate decryption keys on successful attestation via KMS policies bound to enclave measurements. (docs.aws.amazon.com)
For cross‑party comparisons (e.g., mall watchlist vs. tenant list), use Private Set Intersection (PSI) to only reveal matches. (github.com)

Provenance and transparency logs

Record signed manifests (C2PA evidence, EAT claims digests, pipeline configs) to an append‑only transparency log (e.g., Sigstore Rekor v2) and/or IETF SCITT‑style service for auditability and cross‑domain sharing. (blog.sigstore.dev)

Data availability and anchoring

Anchor Merkle roots of daily manifests to the DA layer appropriate to retention and cost: Ethereum EIP‑4844 blobs (ephemeral ~18 days), Celestia (DAS + fraud proofs), Avail (KZG‑backed DAS), or EigenDA (operator DA, high throughput). (ethereum.org)

Compliance, governance, and lifecycle

Emit VC‑backed audit logs to satisfy AI Act record‑keeping (lifetime logging + ≥6‑month retention) and CPPA ADMT risk documentation. Map governance to NIST Privacy Framework 1.1 updates. (ai-act-law.eu)

Pattern 1: “Authenticity at the lens” for video evidence

Goal: From shutter to court or regulator, prove the who/what/when/where of footage without exposing identities unless necessary.

Capture-time signing: Enable Content Credentials (C2PA) on supported cameras (e.g., Leica M11‑P ships with built‑in CC; Sony Alpha series enables C2PA via firmware and a Camera Authenticity Solution). This embeds cryptographic provenance, edit history hooks, and—on Sony—sensor‑based 3D depth signals that help distinguish photographed scenes from re‑shots of screens. (blog.adobe.com)
Device attestation: Where possible, include an EAT in the capture session to assert device model, firmware, secure element status, and nonce‑based freshness. (ietf.org)
Provenance manifests: Write a compact, signed manifest per clip/photo (hashes; C2PA claimset digest; EAT digest; capture geofence policy ID). Push the manifest to a transparency log (Rekor v2) and pin an inclusion proof on a DA layer checkpoint. (blog.sigstore.dev)
Selective disclosure: When sharing evidence, present a VC 2.0 package that reveals only required fields (time window, location grid cell, device attestation OK flag) using BBS+ or SD‑JWT VC presentations. (w3.org)
Verifier UX: Provide a one‑click validation flow that checks C2PA signature chain, EAT verification, and log inclusion proofs; show chain‑anchored timestamps (e.g., day‑batch Merkle root anchored to Celestia or Ethereum). (celestiaorg.github.io)

Why it works:

Cryptographic linking across C2PA → EAT → transparency log → DA anchor establishes tamper‑evidence without over‑exposing identities.
Courts/regulators get independent verification paths and short proofs.

Gotchas:

Ensure CC/C2PA private keys rotate and revocation lists are distributed; C2PA v2.2 refines trust lists and time‑stamping—use it. (c2pa.org)
Some platforms strip metadata; your manifest + log proofs must still verify authenticity even if EXIF/XMP is removed. (c2pa.org)

Pattern 2: Private watchlist matching with TEEs and PSI

Scenario: A stadium operator compares live face embeddings to a law‑enforcement watchlist while minimizing personal data exposure.

Enclave processing: Stream embeddings to a Nitro Enclave. The enclave attests (CBOR/COSE doc, signed by Nitro PKI) and only then unwraps comparison keys from KMS based on PCR values (e.g., ImageSha384, PCR0/1/2). (docs.aws.amazon.com)
PSI for cross‑party comparisons: If a third party contributes a list (e.g., banned patrons), perform PSI so each side learns only the intersection. Mature open‑source PSI protocols (ECDH, Bloom filters/GCS) cover set cardinality and exact matches. (github.com)
Oblivious transport: Submit enclave outputs via OHTTP to decouple request metadata (IP, location) from payload contents, reducing linkability across sessions. (ietf.org)
Minimal proofs: Issue a Verifiable Credential that attests “Face X matched watchlist policy Y at T with enclave Z attested”; avoid disclosing the raw face embedding or non‑matches. Prefer BBS+ or SD‑JWT VC. (w3.org)
Auditing without leakage: Log only derived proofs and enclave attestations into Rekor/SCITT—never the full embedding vectors. (blog.sigstore.dev)

Compliance alignment:

AI Act logging (events over system lifetime; ≥6‑month retention) is satisfied by append‑only logs and DA anchors; CPPA ADMT documentation can reference enclave policies and PSI protocol settings. (ai-act-law.eu)

Pattern 3: Cost‑right data availability and anchoring strategy

Your anchoring choice depends on retention, queryability, and independence.

Short‑lived/price‑sensitive: Ethereum blobs (EIP‑4844). Post daily manifest roots to blobs to cut costs dramatically; blobs are pruned after ~18 days by design. Use a cold archive for long‑term raw storage (e.g., S3/object store) and maintain independent inclusion proofs. (ethereum.org)
Modular DA (probabilistic verification): Celestia offers data availability sampling (DAS) + fraud proofs; light clients sample erasure‑coded shares and can reject bad encodings via fraud proofs. Good for public transparency with light‑client verification. (celestiaorg.github.io)
Modular DA (succinct verification): Avail uses KZG commitments with DAS, enabling strong availability guarantees with few samples and no fraud proofs—useful if verifier devices are constrained (e.g., mobile). (docs.availproject.org)
Operator DA for high throughput: EigenDA reports 100 MB/s write throughput with ~5s avg latency in 2025 V2—useful when anchoring many manifests per minute (e.g., city‑wide sensor grids). (blog.eigencloud.xyz)

Tip: Keep the on‑chain/DA payload to a Merkle root plus minimal metadata (timestamp, policy version). Your transparency log (Rekor/SCITT) remains the primary corpus; DA anchors provide global, independently verifiable checkpoints. (blog.sigstore.dev)

Implementation details that save months

Identity and issuance:
- Use VC 2.0 data model for device/org credentials; standardize issuance via OID4VCI 1.0 (now progressing to Final spec in 2025). (w3.org)
- Prefer DID methods you can resolve offline (did:key, did:web for pilots; migrate to method with strong rotation/recovery for production). (w3.org)
- For selective disclosure, BBS+ (W3C Data Integrity BBS) and SD‑JWT VC both have active test suites and implementer momentum—avoid proprietary formats. (w3.org)
Capture/authenticity:
- Enable C2PA at point of capture where hardware supports it (Leica; Sony firmware with authenticity licenses); design fallback “edge signer” for legacy cameras. (blog.adobe.com)
- Adopt C2PA 2.2 behaviors (time‑stamps, revocation info in update manifests, trust list EKU constraints) in your validators. (c2pa.org)
Attestation and TEEs:
- Model attestation using IETF RATS roles (Attester, Verifier, Relying Party). Emit EATs as JWT/CWT depending on footprint. Gate keys on attested PCRs. (ietf.org)
Transport/minimization:
- Add OHTTP relays for telemetry and credential presentations to prevent server‑side linking of requests. (ietf.org)
Transparency and auditing:
- Use Rekor v2 for cheaper, tile‑backed verifiable logs and routine monitoring (auditors check append‑only consistency proofs). (blog.sigstore.dev)
- For multi‑ecosystem assertions (e.g., OEM → integrator → operator), consider SCITT for signed statement transparency and interop patterns. (datatracker.ietf.org)
Governance:
- Align privacy governance with NIST Privacy Framework 1.1 (draft updated April 14, 2025) to harmonize with CSF 2.0 and AI/ADMT risk themes. (nist.gov)

Regulatory map: how the architecture proves compliance

EU AI Act (high‑risk):
- Art. 12 requires lifetime event logging; Art. 19 requires keeping auto‑generated logs ≥6 months. Use append‑only transparency logs and DA anchors to prove integrity, and VC‑backed evidence packets to support audits or incident reconstruction. (ai-act-law.eu)
California CPPA (ADMT, audits, risk assessments):
- Maintain model/data usage records, attestation proofs for compute environments, and selective‑disclosure dossiers for user rights requests; new rules effective Jan 1, 2026 define timelines for audits/assessments. (cppa.ca.gov)
Chain of custody:
- C2PA Content Credentials plus EAT and Rekor provide multi‑party‑verifiable provenance; DA anchors add independent, time‑stamped checkpoints. (c2pa.org)

Performance and cost notes (2025 reality)

Anchoring costs: EIP‑4844 blobs drastically reduce data posting cost for rollups and commitment anchoring; design for blob pruning after ~18 days—rotate anchors and archive raw data off‑chain. (ethereum.org)
Throughput scaling: When anchoring many manifests per minute, operator DA (EigenDA V2) shows 100 MB/s sustained writes with ~5s average latency; validate SLA needs before choosing. (blog.eigencloud.xyz)
Light‑client verification: Celestia’s DAS allows cheap verification by many light clients; Avail’s KZG approach allows strong guarantees with few samples—useful for mobile verifiers. (celestiaorg.github.io)

Common pitfalls (and how to avoid them)

Treating C2PA as “just metadata”: Enforce trust lists, time‑stamp and revocation validation per v2.2; don’t rely on EXIF alone. (c2pa.org)
Skipping device/compute attestation: Without EAT/TEE proofs, adversaries can replay or inject tampered streams. Build policies that gate decryption on verified PCRs and verifier policy results. (ietf.org)
Over‑sharing during verification: Adopt selective disclosure (BBS+/SD‑JWT VC) so verifiers learn only the minimal facts. (w3.org)
“On‑chain everything”: Anchor proofs only; keep raw media off‑chain. Use Rekor/SCITT for searchable provenance, DA for checkpoints. (blog.sigstore.dev)
Misunderstanding blob retention: Ethereum blobs are pruned (~18 days); set rotation/archival policies or choose a DA layer with different retention. (ethereum.org)

Procurement checklist (copy‑paste into your RFP)

Cameras/sensors:
- Supports C2PA capture‑time signatures; publishes key rotation and revocation procedures; optional device EAT support. (c2pa.org)
Analytics platform:
- Offers TEE execution with verifiable attestation; key release tied to PCRs; PSI for cross‑party matching; OHTTP support. (docs.aws.amazon.com)
Identity and credentials:
- Issues VC 2.0 credentials via OID4VCI; supports BBS+ and SD‑JWT VC presentations; supports DID methods with rotation/recovery. (w3.org)
Auditability:
- Writes to a verifiable transparency log (Rekor v2/SCITT); provides inclusion/consistency proofs and retention policy mapping to AI Act/CPPA. (blog.sigstore.dev)
Anchoring:
- Configurable DA target (Ethereum blobs, Celestia, Avail, EigenDA) with documented costs, retention, and light‑client verification story. (ethereum.org)
Governance:
- Provides NIST Privacy Framework mapping (v1.1) and change‑control for models/policies. (nist.gov)

A brief example: city‑scale incident reconstruction with privacy

Setup:
- Intersection cameras sign frames with C2PA; gateway runs in Nitro Enclave, attested before decrypting frame hashes. Manifests (hashes, EAT digest) go to Rekor v2 hourly; daily Merkle root anchored to Celestia. (sony.mediaroom.com)
Event:
- After an incident, investigators request a VC package proving “vehicle ABC123 seen between 21:10–21:20 in zone G.” The holder presents a minimal disclosure proof (time window, zone, signed by operator, anchored on day‑root). (w3.org)
Privacy:
- Faces/plates not in the query never leave enclaves; PSI used to compare sightings with stolen‑vehicle list; auditors verify logs for AI Act compliance and six‑month retention. (github.com)

Emerging practices to adopt now

Emit single‑file “evidence bundles” that contain: C2PA validations, EAT verification result, Rekor inclusion proof, DA anchor reference, and a VC with selective disclosures.
Monitor transparency logs automatically; Rekor v2 lowers ops cost—set up external witnesses/monitors. (blog.sigstore.dev)
Standardize issuance and verification flows with VC 2.0 + OID4VCI; avoid proprietary attestation formats—stick to RATS/EAT. (w3.org)

Final take

You no longer have to choose between audit‑grade provenance and privacy. With capture‑time C2PA, RATS/EAT device and compute attestation, selective‑disclosure VCs, verifiable transparency logs, and modular data availability, you can design surveillance architectures that regulators can verify and the public can trust—without centralizing sensitive footage on‑chain or over‑sharing identities. The building blocks are standardized, interoperable, and production‑ready.

If you’re piloting, start small: one capture device class with C2PA enabled, one enclave‑based analytic, Rekor v2 for provenance logs, and a daily anchor on Celestia or Ethereum blobs. Expand from there with OID4VCI issuance, BBS+/SD‑JWT VC presentations, and cross‑party PSI when collaboration begins.

7Block Labs can help you turn this into a concrete delivery plan: device onboarding, attestation policies, VC schema design, and DA anchoring that fits your risk and budget profile.