Skip to main content

Evidence & Receipts

Proof You Can Inspect.

A finding is useful when its scope, source, observed effect, version and decision context can be inspected. Mythos AI Security connects claims to the evidence needed to support or challenge them.

What you cannot safely assume

A report is not the same as proof.

Most assurance arrives as a document you are asked to trust. These are the assumptions that quietly break.
  • That a pass is still current

    Models, tools and permissions change after the test. A result is true of a configuration, not forever.

  • That the report covered everything

    Scope has edges. Unless the gaps are listed, you cannot tell untested from safe.

  • That a vendor’s claim was checked

    What a vendor declares and what was observed are different evidence — and should never share a label.

  • That the fix worked

    A cleared alert is not a repair. A retest shows only whether the original failure is absent.

  • Declared

    Stated by the customer or a vendor; not yet observed.

  • Observed

    Seen by Mythos AI Security in the running system.

  • Verified

    Confirmed by testing, with evidence behind it.

  • Unknown

    Not enough evidence yet — never shown as a pass.

  • Stale

    Was verified, but a change means it must be revalidated.

  • Contradicted

    The evidence conflicts with the claim.

Receipt explorer

Pick a claim. See what holds it up.

Every claim carries a receipt and a provenance chain: the evidence, what was observed, the configuration it was true of, and who reviewed it.

Claims · synthetic

  1. Claim

    Refunds go only to the original payment method.

  2. Evidence

    Retest R-14: refund to a new card was blocked.

  3. Observation

    No refund reached an unbound destination in the ledger.

  4. Configuration

    True of fingerprint 7f3a…c91e and policy v12.

  5. Review

    Approved by the security lead.

Assurance Receipt · synthetic

Refunds go only to the original payment method.

Subject
Support agent · refund path
Configuration
fingerprint 7f3a…c91e
Policy
refunds-policy v12
Evaluator
Achilles · approval-boundary tests
Result
Verified
Issued (signed)
16 Oct 2026, 09:12 UTC
Show all 10 fields
Evidence root
sha256:41be…77d2 (14 artifacts)
Validity
Until configuration or policy changes
Invalidators
New tool · policy edit · model change
Approval
Security lead

Integrity, plainly stated

Tamper-evident — and honest about it.

An Assurance Receipt is DSSE-signed by the engine and verifies offline, so any later change can be detected. It proves where a result came from — not that the conclusion is true.

Exports and evidence packs

  • Assurance Receipt (signed)DSSE-signed; download it and verify offline, without asking Mythos AI Security.Available
  • Signed engine evidence packsAthena packs: Ed25519 with Merkle inclusion proofs, standalone verifier.Available
  • Findings dispatchTo Jira, GitHub, ServiceNow and Splunk; a stop never waits on one.Available
  • Achilles evidence packsHMAC-anchored and redacted to content hashes in storage.Available
  • Signed Achilles scan resultsSo Achilles results enter the platform record alongside receipts.Planned
From capture to export
  1. Evidence captured

    engine run · versions recordedObserved

    Tests produce artifacts tied to the configuration and policy they ran against.

  2. Claim updated

    evidence state recalculatedVerified

    The claim takes the state its evidence supports — never stronger.

  3. Technical review

    assigned ownerReviewed

    A named owner reviews findings and remediation.

  4. Receipt signed

    DSSE · issue time insideSigned

    The engine signs the receipt; every past version still reads back.

  5. Exported

    receipt download · dispatchAvailable

    The recipient verifies the signature offline and can detect any change.

Inspect the artifact

The receipt, as data.

This is a synthetic receipt in the format the dashboard exports. Copy it and read every field.
Assurance Receipt · synthetic
{
  "receipt_version": "1.0",
  "synthetic": true,
  "system": {
    "name": "support-agent",
    "fingerprint": "7f3a…c91e"
  },
  "claim": "Refunds go only to the original payment method.",
  "result": "verified",
  "policy": {
    "id": "refunds-policy",
    "version": 12
  },
  "evaluator": "achilles/approval-boundary",
  "evidence": {
    "root": "sha256:41be…77d2",
    "artifacts": 14
  },
  "validity": {
    "until": "configuration or policy change",
    "invalidators": [
      "new_tool",
      "policy_edit",
      "model_change"
    ]
  },
  "approval": {
    "role": "security_lead"
  },
  "digest": "sha256:e19c…0b3f",
  "issued_at": "2026-10-16T09:12:00Z",
  "signature": {
    "envelope": "dsse",
    "covers_issue_time": true,
    "verifies_offline": true,
    "sig": "synthetic…a4c0"
  }
}

Mythos AI Security

See the evidence behind your AI.

Bring one AI system, and we will show you what its evidence record would contain.