Skip to content

Changelog

Released versions of the untrust CLI and scanner. The hosted console is updated continuously; its changes appear in the console itself.

Changes in current builds after 1.3.0

Design-partner wheels are built from the current product source and keep the version number 1.3.0 until the next numbered release. These behaviour changes apply to a wheel built after 1.3.0:

  • One exit-code contract for every offline verifier. untrust ci verify, untrust ledger verify and untrust pack verify exit 0 verified, 1 not verified, 2 a required input is missing, 3 unsigned or unbound. Details.
  • Evidence packs are bound to their organization. untrust pack verify requires --organization, and untrust ci verify requires --public-key and --org. Packs that name no organization never verify: generate new ones.
  • API keys follow their creator. A key's permissions are capped by its creator's current role on every request. Automation whose key creator has left must be re-keyed by a current administrator.

1.3.0 (2026-10-17)

First design-partner build, distributed as a reproducible wheel with a SHA-256 digest and a CycloneDX SBOM (Install the CLI).

1.2.0 (2026-09-30)

Third platform: NVIDIA GPU Confidential Computing.

Added

  • NVIDIA GPU CC platform (--platform gpu-cc, 18 checks) across both TEEs of a confidential-GPU deployment: report trust, verification, attestation ordering, model inputs, memory protection, the confidential VM and its binding to the GPU, and key and identity bypasses.
  • GPUCC-SIGVERIFY-01 checks a retained nvattest result, including that its nonce matches the one it ran with, so an old passing result cannot be replayed.
  • GPUCC-CUDA-01 proves a CUDA kernel ran on the attested GPU after a successful attestation.
  • Assurance tiers on every GPU CC check (report-derived, probed, operator-declared).
  • GPU CC target flags (--gpu-attestation-report, --gpu-verifier-policy, --gpu-cc-mode, --gpu-kbs-policy, --gpu-model-bucket, --gpu-launch-config).

Fixed

  • --read-only applies on every platform: it also skips the canary-write probes GCS-BOOT-02 and GPUCC-MODEL-01.
  • --platform no longer offers tdx, which had no checks.
  • untrust --version reports the package version.
  • GPUCC-RIM-01 fails a report that omits a required measurement instead of skipping it, and GPUCC-KEY-01 passes only an attestation-gated key broker as the key source.

1.1.0 (2026-08-30)

Second platform: AMD SEV-SNP on GCP Confidential Space.

Added

  • Confidential Space platform (--platform sev-snp, 11 checks): attestation-bound key release through the Workload Identity Federation attribute condition, Cloud KMS federation, bootstrap-bucket hardening with an active injection probe, Confidential VM configuration, attestation token claims, and five attestation-bypass checks (metadata mutability, service-account impersonation, service-account keys, sibling WIF providers and workload image signing).
  • --platform {nitro,sev-snp}, GCP target flags and platform-aware list-checks.

Changed

  • The GCP client libraries are core dependencies, so one install provides both the AWS and GCP scanners.

1.0.0 (2026-07-30)

First release.

Added

  • 33 AWS Nitro Enclaves checks across bootstrap and supply chain, object storage, KMS attestation, IAM and access control, network and host, host memory and persistence hygiene, enclave runtime, and detection and forensics.
  • Alternative state-store checks: DynamoDB (DDB-01), Secrets Manager (SECRETS-01), SSM Parameter Store (SSM-01), EFS/EBS (EFS-01) and RDS/Aurora (RDS-01).
  • --read-only to run passive cloud-API checks only.
  • untrust scan and untrust list-checks; --demo; JSON output with --output and --json.

Fixed

  • KMS-01 resolves key aliases before reading the key policy.
  • BOOTSTRAP-01 treats any client-side rejection of the write probe as blocked (PASS); only server-side or throttling failures are ERROR.
  • Customer-managed key detection for state-store checks uses kms:DescribeKey rather than an alias-name heuristic.