Adverserial AI Enter platform
ADVERSERIAL AI / PUBLIC TRUST MATERIALVERIFY.ADVERSERIAL.AI

INDEPENDENTLY READABLE EVIDENCE

Verification
registry.

Public policy, source, and release evidence for Adverserial confidential inference. This host is an evidence reader: it never receives prompts, completions, API keys, account data, or model traffic.

ACTIVATION STATUS

Production verification policy.

POLICY PUBLISHED

The CyberSeek policy pins the approved CPU boot measurements, dstack launch identity, inference runtime, model, deployment configuration, TLS key, and receipt signer. Compatible clients compare these pins against fresh hardware evidence before sending a prompt. This registry publishes evidence; it does not perform a live endpoint check.

APPROVED PRODUCTION SNAPSHOT

CyberSeek verification pins.

These are the expected values in the published policy. Clients must verify fresh hardware evidence against them. This static page is not a live health check or a guarantee of model response quality.

Model
adverserial/cyberseek
Policy published
2026-11-18T17:20:52.000Z
Policy SHA-256
sha256:eaf1cf25104029c3dabd1e2284c3b23cbab0703eae1401e74e62a354d79a0ec7
Model artifact
sha256:bd18db14b7f1e98b93401d194d62c7e3b6282c7751ae1a9d523079b03e84c90e
Inference runtime
sha256:b4f09a2ccbcfcc262d49600c3cd7f8581a793ea7d104947257b6a95500953b90
Deployment configuration
sha256:f32d9b51afc418c0b7151da55d517f2b5b3354f7cd8e424c0edd2a3ed0d5cbdd
TLS public key (SPKI)
sha256:zD557zx8m2gI67trOhh-ajvJlyfHMgv_U_wrV8TJWMw
Receipt signer (ES256)
mAI2xsXKerRhngOm15IT4f41rE_Cpgp0G3szP8YwYdk
Proxy release
v0.1.0-rc.29
TDX mrtd
b24d3b24e9e3c16012376b52362ca09856c4adecb709d5fac33addf1c47e193da075b125b6c364115771390a5461e217
TDX rtmr0
d8f547a53863772b1a8b64f6b9ae70ad98549fbb44628d2f07c8ed6f883ce6288611538b27dac0eb5baa216a4154a8bd
TDX rtmr1
07e6f51aa763abfe75c3ddfbf4f425fe3f0ceff66d807a75e049303dce9addf68e7218729bd419638af63a370f65878c
TDX rtmr2
f1c82667a354194467cd8419efd14a714560dd9b85b4c13b25c11e44bf4e126248c2255fad58c303fb0ca2921765d53a
TDX app_id
b74e3dde6292cc69bac759b22fc723a299aa8a83
TDX compose_hash
bd773fb41fd23cee2aa255b96403d7d7cbba90785ab1cc4513ebaa26db4a815b
TDX os_image_hash
806a352e16175d90568de97dff563f31f680239e6b90e9b5b2e9141d0955b0d9

Verify now: ack model adverserial/cyberseek then ack attestation. Policy distribution uses HTTPS; the policy has no detached signature.

01 / REGISTRYPUBLIC JSON

Policy registry

Current model: CyberSeek. Deployed inference proxy: v0.1.0-rc.29. Latest published proxy tag: v0.1.0-rc.29. Proxy release numbers are separate from model and inference-engine versions. Earlier manifests remain in the release archive.

02 / SOURCEPUBLIC REPOSITORIES

Auditable source

The proxy, client SDK, and deployment definitions are reviewed and released separately from this site.

ACTIVATION GATE

What must be true
before a client sends data.

  1. 01

    Release image digests and GitHub build provenance are published. Verify the provenance signatures against the expected repository and workflow.

  2. 02

    The HTTPS-published policy pins model, runtime, configuration, boot measurements and receipt keys. It has no detached policy signature.

  3. 03

    The client verifies a fresh nonce-bound Intel TDX quote and NVIDIA evidence.

  4. 04

    The quote binds the endpoint TLS, encrypted-body receiver, and receipt keys. Clients replay RTMR3 and compare the measured workload against the published policy.

  5. 05

    Clients enforce receipt expiry and verify ES256 inference receipts. Attestation receipts expire after 5 minutes. NVIDIA evidence has a separate maximum age of 30 minutes.

WHY THIS HOST IS SEPARATE

Evidence stays public.
Customer data does not.

Verification material can be fetched and audited without contacting the inference runtime. The confidential client uses this evidence to make a local trust decision before it creates an encrypted connection to an attested endpoint.

Read the confidential architecture