Privacy Preserving Verifiable Geofencing with Residency Proofs for Sovereign Workloads

1.1. Scope and Layering

V-GAP covers platform integrity verification (Layer 2) and residency verification (Layer 3). It assumes that the binding of individual workloads to the local Workload Identity Agent (Layer 1) is established by a co-location mechanism such as {{!I-D.mw-wimse-transitive-attestation}}. Together, these three layers form a complete chain of trust from silicon to workload, as required by the WIMSE Architecture {{!I-D.ietf-wimse-architecture}}.¶

1.2. Relationship to Related Work

defines how a Verifier encodes geographic location conclusions — jurisdiction-level results such as country, subdivision, and city — as EAT Attestation Result claims for consumption by a Relying Party. That draft addresses the output encoding side of the attestation pipeline.¶

V-GAP addresses the complementary input side: the Evidence profile the Attester produces, the hardware-binding mechanism (TPM quote) that makes location evidence verifiable, and the verification procedure the Verifier applies to produce an Attestation Result. V-GAP Evidence is what a Verifier appraises to yield the kind of geographic result claims that {{I-D.richardson-rats-geographic-results}} encodes.¶

The two documents are intended to compose: a Verifier that appraises a V-GAP lah-bundle could express its conclusion as an Attestation Result using the geographic claims defined in {{I-D.richardson-rats-geographic-results}}. V-GAP is self-contained; use of that encoding is OPTIONAL and is one possible way a Verifier may express its conclusions — consumers MAY enforce geofence policy directly from the Attestation Result, use the V-GAP X.509 extension (OID 1.3.6.1.4.1.65284.1.1) as the trust signal, or adopt any other result encoding their deployment requires.¶

One gap in the combined stack is not addressed by either document: the mapping from a raw location fix or geofence proof to a named legal jurisdiction (for example, from "inside polygon P" to "in jurisdiction X"). This mapping raises its own trust questions — who maintains the polygon-to-jurisdiction database, under what authority, and how that mapping is kept current — and is deferred to future work.¶

2.1. Abbreviations

• AK: Attestation Key¶
• BMC: Baseboard Management Controller¶
• DAA: Direct Anonymous Attestation¶
• EAT: Entity Attestation Token¶
• EK: Endorsement Key¶
• GNSS: Global Navigation Satellite System¶
• IMA: Integrity Measurement Architecture¶
• IMEI: International Mobile Equipment Identity¶
• IMSI: International Mobile Subscriber Identity¶
• LAH: Location Anchor Host¶
• OOB: Out-of-Band¶
• PCR: Platform Configuration Register¶
• PoR: Proof of Residency¶
• SPDM: Security Protocol and Data Model¶
• STARK: Scalable Transparent ARgument of Knowledge¶
• SVID: SPIFFE Verifiable Identity Document¶
• TEE: Trusted Execution Environment¶
• TPM: Trusted Platform Module¶
• V-GAP: Verifiable Geofencing Attestation Profile¶
• ZKP: Zero-Knowledge Proof¶

4.1. Server-centric Enforcement

Enterprises need cryptographic proof that workloads run only on approved hosts within an approved geographic boundary, and that data flows only between approved boundaries.¶

• Workload-to-workload (general): Relying parties accept workload identities only when the issuing host attests platform integrity and "in-zone" residency, preventing credentials from being used outside the approved boundary.¶

• Agentic AI workloads: An AI agent may access sensitive data or perform sensitive actions only when its Workload Identity Agent presents hardware-rooted integrity evidence and a verifiable "in-zone" proof (optionally privacy-preserving), binding identity to both platform state and residency.¶

• Federated / edge AI (key or model release): High-value artifacts (e.g., decryption keys or model weights in federated learning) are released only when the partner/edge host attests it is integral and resident within the required boundary. This is useful for intermittently connected sites.¶

• User-to-server: Clients validate that the server endpoint is operating within an approved boundary (e.g., by policy tied to the server's attested identity and residency evidence).¶

4.2. User-centric Enforcement

Enterprises may also need trustworthy location signals for user-facing access decisions.¶

• Geofenced access control: User access is permitted only when the user (or user device) proves it is within an allowed boundary, ideally without requiring precise location disclosure.¶

• On-premises boundaries: Customer-premises equipment can define an enterprise boundary, with a network or enterprise infrastructure providing supporting evidence for policy enforcement.¶

• Restricted support geographies: Administrative or support actions can be allowed only when the operator proves presence within allowed geographies, reducing policy and insider-risk exposure.¶

4.3. Compliance and Risk Reduction

Geofence attestation provides audit-ready evidence to support data residency and sovereignty controls, and it can also reduce non-compliance risk from misconfiguration or spoofable signals. Even when not mandated, "in-zone" proofs help address: configuration drift, edge relocation/proxying, contractual residency requirements, and location-privacy minimization (proving "inside the zone" without storing coordinates).¶

5.1. RATS Role Mapping

V-GAP instantiates the RATS Architecture {{!RFC9334}} with the following role assignments.¶

Note 1: "Downstream Relying Party" is not a role defined by {{!RFC9334}}. It is used here to distinguish the entity that consumes the issued credential from the Relying Party that consumes the Attestation Result.¶

5.2. Evidence Flow

The V-GAP evidence flow follows the RATS background-check model ({{!RFC9334}}, Section 3.2): the Attester conveys Evidence to the Verifier, the Verifier appraises it and conveys the Attestation Result to the Relying Party.¶

Attester (LAH)
    |
    | V-GAP Evidence (lah-bundle)
    v
Verifier (Host Identity Mgmt Plane)  <--- Endorsement (MNO)
    |
    | Attestation Result
    v
Relying Party + Credential Issuer (Workload Identity Mgmt Plane)
    |
    | X.509-SVID with V-GAP extension
    v
Downstream Relying Party (mTLS peer)

The Relying Party in V-GAP also acts as a Credential Issuer (CA): it materializes the Attestation Result into an X.509 workload identity credential (for example, a SPIFFE SVID) containing the V-GAP Evidence as a CRITICAL extension. This "trust translator" pattern allows downstream consumers to rely on standard X.509/mTLS verification without needing to understand RATS or V-GAP directly.¶

Implementations of this profile MUST mark the X.509 extension containing the V-GAP Evidence Bundle as CRITICAL. When marked CRITICAL, any downstream Relying Party that does not understand the extension MUST reject the credential, enforcing fail-closed behavior for residency-constrained workloads.¶

5.3. V-GAP Evidence Structure

The lah-bundle is the RATS Evidence structure defined by this profile. It is a hardware-sealed object embedded as an X.509 extension (OID 1.3.6.1.4.1.65284.1.1) in the workload identity credential (for example, a SPIFFE SVID). It binds a workload identity to physically verifiable claims — TPM hardware identity, privacy-preserving geolocation, and agent binary integrity — without exposing PII.¶

{
  "lah-bundle": { },
  "mno-endorsement": { },
  "workload": { }
}

5.4. TPM Quote Verification Procedure

The Verifier MUST perform the following steps to validate the tpm-quote-seal:¶

1. Decode tpm-quote-seal (Base64URL → bytes)¶
2. Parse TPMS_ATTEST structure¶
3. Assert TPMS_ATTEST.type == TPM_ST_ATTEST_QUOTE¶
4. Compute expected_qd = SHA-256(JCS({tpm-ak, geolocation-id-hash, geolocation-proof-hash, privacy-technique, nonce, timestamp, workload-identity-agent-image-digest}))¶
5. Assert TPMS_ATTEST.qualifyingData == expected_qd¶
6. Verify signature over TPMS_ATTEST bytes using tpm-ak public key (RSASSA-PKCS1-v1_5 or ECDSA)¶

If any step fails, the Verifier MUST reject the Evidence and MUST NOT produce a positive Attestation Result.¶

5.5. Freshness and Replay Prevention

To prevent mix-and-match and replay attacks, Verifiers MUST enforce the following:¶

• Attestation Results MUST be fresh and MUST be bound to the credential issuance event (for example, by cryptographically binding freshness values used for platform quotes and workload credential issuance within the Verifier result).¶
• The nonce field in the lah-bundle MUST be a freshness value issued by the Relying Party (Workload Identity Management Plane) for each attestation interval.¶
• Verifiers MUST reject Evidence where the timestamp falls outside the configured freshness window.¶

Where policy requires it, the Verifier can additionally require that an agent software measurement (for example, image digest) is covered by validated platform Evidence, reducing the risk that a modified or unauthorized agent obtains credentials.¶

6.1. Location Spoofing

6.2. Zero-Knowledge Proof Security

V-GAP's privacy-technique = "zkp" mode uses Plonky2 STARK proofs. The following properties and threats apply:¶

• Circuit correctness is the primary attack surface. A ZKP proves only what its arithmetic circuit encodes. Errors in the geofence boundary circuit — including precision errors, off-by-one boundary conditions, or incorrect coordinate system handling — yield proofs that are cryptographically valid but semantically incorrect. The geofence circuit MUST be independently audited before deployment.¶

• No trusted setup. Plonky2 is STARK-based and requires no trusted setup phase, eliminating the class of attacks arising from compromised SNARK setup parameters. Implementations substituting a different zkp-format MUST ensure it also provides transparent setup, or MUST document the resulting trust assumptions.¶

• Computational soundness. STARK security is computational, not unconditional, and relies on the collision resistance of the underlying hash function. Implementations SHOULD target at least 128-bit security and MUST document the proof system parameters (field size, hash function, FRI parameters) to enable independent security analysis.¶

• Proof freshness. A valid ZKP proves location at proof-generation time. The nonce and timestamp freshness requirements that apply to tpm-quote-seal apply equally to ZKP proofs: Verifiers MUST reject proofs whose timestamp falls outside the configured freshness window.¶

• Prover integrity. A compromised prover can produce false proofs even for a correctly specified circuit. This threat is mitigated by V-GAP's TPM binding: the tpm-quote-seal covers geolocation-proof-hash, so a false proof can only be embedded in a bundle that also passes TPM quote verification for an approved platform. The ZKP privacy guarantee is only meaningful in conjunction with verified platform integrity.¶

B.1. Gating Credentials on Verified Evidence

This profile assumes two cooperating control planes, mapped to RATS roles:¶

• Verifier (Host Identity Management Plane): Appraises platform integrity and residency Evidence and produces an Attestation Result.¶
• Relying Party + Credential Issuer (Workload Identity Management Plane): Issues or renews workload identities (for example, SVIDs) only when the Attestation Result satisfies policy. Also acts as the CA that signs the resulting credential.¶

In intermittently connected edge deployments, local operation can continue during outages, while centralized policy can be enforced on renewal and on release of high-value secrets once connectivity is available.¶

B.2. Distributed Identity Issuance and Scaling

To support edge deployments and intermittent connectivity, identity issuance may be distributed within a sovereign boundary.¶

• Edge issuance: Workload identities (for example, SVIDs) may be issued by an issuer deployed within the same boundary as the workloads.¶
• Scoping: Issued identities should be scoped so they are not accepted outside the intended deployment boundary (for example, via trust bundle partitioning and policy).¶
• Renewal gating: Issuers should renew short-lived identities only when the Verifier result for integrity and residency is valid for the requested freshness window.¶

B.3. Mobility and Sovereign Handover

When a workload moves between anchors or boundaries, the Workload Identity Agent should obtain a new V-GAP bundle that reflects the new LAH and current residency.¶

Verifiers should treat this as a normal re-attestation event: - platform integrity continuity can remain stable, but - residency checks should be re-evaluated for the new anchor/boundary.¶

B.4. Location Anchor Hosts

To scale location sensing, a deployment may use dedicated anchors:¶

• End-user anchors: A user device (for example, a phone) can serve as an LAH for a nearby client device. The mechanism by which the anchor establishes its own location (and any proximity evidence it may provide) is out of scope for this document.¶
• Data center anchors: A small set of hosts can act as LAHs for a cluster. Timing-based mechanisms (for example, PTP-derived) may assist in establishing relative location; protocol details are deferred to future profiling work (see {{I-D.ramki-ptp-hardware-rooted-attestation}}).¶

C.1. Nonce Chain and Merkle Audit Log

One way to satisfy the freshness requirements in this profile is through a chained nonce and Merkle audit log. Where bundle[n] denotes the JCS-canonicalized lah-bundle object at attestation interval n:¶

chain[n] = SHA-256(chain[n-1] || SHA-256(JCS(bundle[n])))
nonce[n]  = HMAC(secret, n || chain[n-1])

C.2. Key Registry and Synchronization

• A Cloud (central) Verifier (Host Identity Management Plane) maintains a registry of accepted AK public keys and associated metadata (for example, EK certificate chain, hardware identity, and status).¶
• An Edge Verifier may maintain a local registry to support disconnected operation and periodically synchronizes updates to the central registry.¶

C.3. Key Rotation

To prevent rogue key injection during rotation:¶

• The central registry should accept a new AK only if the edge plane provides a rotation proof that chains the new AK to previously accepted state.¶
• A rotation proof should be a JCS-canonicalized object signed by the previously accepted AK (or, if available, validated by a fresh hardware-rooted OOB quote).¶

{
 "new-ak-pub": "Base64URL_Encoded_Public_Key",
 "serial-number": "AK_Serial_XYZ",
 "timestamp": 1708845600,
 "hardware-uuid": "Host_Hardware_UUID",
 "signature": "Base64URL_Signature_from_Previous_AK"
}

C.4. Credential Activation and Re-Verification

Credential activation (for example, TPM2_MakeCredential) is expensive to run on every request. Verifiers should perform it on events such as:¶

• Initial onboarding
  ¶
• Reboot / reset detection (for example, TPM clock/reset counters)
  ¶
• Policy violations or drift signals (for example, firmware or inventory changes)
  ¶
• Failure of location evidence checks
  ¶
• Explicit elevation to higher assurance policy
  ¶

Between full activations, Verifiers may accept fresh quotes from registered AKs as proof of continued compliance, subject to policy.¶

C.5. Revocation and Health Signals

• The edge plane should maintain a per-node health signal (for example, tamper, firmware policy violations).¶
• On severe health signals, the Verifier should revoke the relevant AK(s) and reject identities derived from them according to policy.¶

C.6. Disconnected Operation (Leased Identity)

For intermittent connectivity, the Verifier may issue identities with extended validity (a lease) under policy. If a lease is used:¶

• The edge plane should revoke or refuse renewal locally on tamper/drift signals.¶
• The workload should re-attest and satisfy current policy on reconnection before renewal or release of high-value secrets.¶

F.1. Example Instance (privacy-technique = "zkp")

{
  "lah-bundle": {
    "tpm-ak": "-----BEGIN PUBLIC KEY-----\nMIIBIjANBgkqhkiG...\n-----END PUBLIC KEY-----",
    "geolocation-id-hash": "7f4a2c1b9e3d8f0a6b5c4d2e1f0a9b8c...",
    "geolocation-proof-hash": "c8bc2ed62a7a650d99e0884197cdf345...",
    "privacy-technique": "zkp",
    "geolocation-payload": {
      "zkp-proof-uri": "https://verifier.example/v1/proof/c8bc2ed6...",
      "zkp-format": "plonky2"
    },
    "nonce": "ZmUyZjdmMzlmZGVlZWQxOTM1YjY0Mjk0...",
    "timestamp": 1740693456,
    "tpm-quote-seal": "ARoAAQALAAUACwEA...",
    "workload-identity-agent-image-digest": "a1b2c3d4e5f6...64-char-hex-sha256"
  },
  "mno-endorsement": {
    "mno-key-cert": "MIIBIjANBgkqhkiG9w0BAQEFAAOCAQ8A...",
    "mno-sig": "MEYCIQDx9z2k..."
  },
  "workload": {
    "workload-id": "spiffe://example.org/python-app",
    "key-source": "tpm-app-key"
  }
}

F.2. Sensor Type Input Recipes

The following recipes define how geolocation-id-hash is constructed from different sensor types. The Verifier sees only the opaque hash — never the raw identifiers.¶