Trust Center

Trust should be designed, not proclaimed.

Controlled access, cryptographic integrity, authority boundaries, human validation and progressive disclosure define Quantum Excellium’s public trust posture.

30-day evaluation licence

Controlled access

Explicit identity, rights and scope before sensitive execution.

Cryptographic integrity

Signatures, hashes, authenticated encryption or attestation according to the threat model.

Human gates

Consequential legal, financial, medical or regulatory workflows are not positioned as autonomous final authority.

Local-first options

Selected systems can prioritize local control, reduced egress and private deployment.

Controlled disclosure

Public pages show positioning; exploit-relevant detail remains in qualified exchanges.

Independent review

Regulated or critical deployments may require external security, legal or compliance review.

Proof, not promises

Verify before you trust

Run the proof

The same AI agent exploited, then governed — 30 seconds, no dependencies: qec-governed-agent-demo. Seven platform scenarios: qec-demo-agents.

Verify any journal

ledger-verify (Apache-2.0): VALID or INVALID, without trusting us. The AI Worlds demos export this format with one button.

QEC 3.5.2

314 automated tests; executable proof scripts run in your environment and produce a signed attestation; delivered packages carry their SHA-256.

Limits, in writing

No SOC 2 or ISO 27001 certification is claimed. QEC provides an integration boundary for the OWASP Agent Control Standard — no conformance claim. No performance figure is published that we have not measured and cannot reproduce for you. Six patent applications filed with INPI (not granted).

The path of one decision under QEC: each stage allows or refuses, and every outcome is written to a chained ledger that a third party verifies without trusting us. The public demonstration code that follows exactly this path is on GitHub (a demonstration shim, not the licensed runtime).

Real outputs, reproduced on 27 September 2026

$ python ledger_verify.py out/governed_ledger.jsonl
VALID

$ sed "5s/[0-9]/7/" out/governed_ledger.jsonl > out/tampered.jsonl   # altère un chiffre de la ligne 5
$ python ledger_verify.py out/tampered.jsonl
INVALID
- line 5: hash mismatch
exit code: 1

$ (cd demo/qec-governed-agent-demo && python -m pytest -q)
8 passed in 0.13s
$ (cd demo2/qec-demo-agents && python -m pytest -q)
6 passed in 0.54s
$ (cd gh/pft-attestation-verify && python -m pytest -q)
7 passed in 0.34s
$ (cd gh/ledger-verify-additions && python -m pytest -q)
6 passed in 0.18s
$ (cd aiwp/ai-worlds-protocol && python -m pytest -q)
9 passed in 0.11s

Standard Python 3.11, no dependencies. Left: the public verifier declares the ledger produced by the demo VALID, then INVALID as soon as one digit of one line is altered, naming the line. Right: the five public proof repositories (36 tests). The QEC 3.5.2 runtime itself is not public; its full suite is reproduced below, and runs on your side under a 30-day evaluation licence.

Licensed QEC 3.5.2 runtime — full suite, executed on 27 September 2026

$ python --version
Python 3.11.15
$ python -m pytest -q
........................................................................ [ 22%]
........................................................................ [ 45%]
........................................................................ [ 68%]
.................................................................. [ 89%]
................................                                         [100%]
314 passed, 6 subtests passed in 4.90s

$ python platform/kernel/prove_trust_plane.py
OK rotation d'ancre : ordre signé par un non-anchor refusé
OK rotation trust anchor : nouvelle racine acceptée
OK rotation trust anchor : ancienne racine refusée
OK révocation de clé durable
OK dérive détectée → exécution interdite
OK état SQLite altéré (révocation effacée) détecté
OK ReplayLab contrefactuel : policy plus stricte aurait bloqué 1 décision sur 2
==> 29/29 scénarios prouvés → qec_trust_plane_proof.json

{
  "overall": "PASS",
  "checks": {
    "pytest_integral": {
      "status": "PASS",
      "passed": 314,
      "skipped": 0,
      "subtests_passed": 6
    },
    "unittest_integral": {
      "status": "PASS",
      "tests": 314
    },
    "trust_plane_3_5_2": {
      "status": "PASS",
      "tests": 43
    },
    "kernel_assurance": {
      "status": "PASS",
      "passed": 10,
      "total": 10
    },
    "trust_plane_proof": {
      "status": "PASS",
      "passed": 29,
      "total": 29
    },
    "differential_matrix": {
      "status": "PASS",
      "verified": 28,
      "delivered": 28
    }
  }
}

Real output of the licensed runtime (not public), standard Python 3.11, no dependencies: 314 tests passed in 4.9 s, expert validator PASS (kernel assurance 10/10, Trust Plane 29/29 adversarial scenarios, differential matrix 28/28 bundles verified). Under an evaluation licence, the same command produces the same output on your side.