VectorCertain
6 ATTACK FAMILIES
HASH-LINKED EVIDENCE CHAIN
SEALED 20-NAME INTERFACE
Certification · Six attack families · Hash-linked evidence

MYTHOS — Adversarial Certification

Every AI safety vendor claims their guardrails work. Almost none can hand you cryptographic evidence of when that claim was last tested, against what, and by whom.

FROZEN-SET EQUALITY VERIFIED ACROSS SPRINTS
STANDING INVITATION: EVALUATORS MAY SUBMIT ATTACK SCENARIOS

MYTHOS is an adversarial certification framework that tests a governance layer against six distinct attack families and emits hash-linked proof of each result. The output is not a marketing assertion — it is an evidence chain a third party can verify without trusting the party that produced it [VC-C].

The design problem it addresses is that assurance decays. A system validated at deployment is not a system validated today: vendors update models the customer does not control, input distributions shift, calibration drifts — and an adversary who can induce overconfidence can push unsafe decisions past a trust threshold without tripping an anomaly alert. Point-in-time validation cannot detect any of that. Continuous adversarial certification can [VC-C].

Request the certification briefing

§ 1 · THE MECHANISMHow Certification Produces Verifiable Evidence

Each certification run exercises the governed deployment against the six attack families and records every result into a hash-linked chain: each entry binds to its predecessor, so no result can be silently removed, reordered, or altered after the fact. Verification requires only the chain and the published interface — not access to MYTHOS internals, and not trust in VectorCertain [VC-C].

The certification surface itself is stability-sealed: MYTHOS ships as a sealed 20-name published interface, verified across subsequent engineering sprints by exact frozen-set equality — no name added, removed, or re-spelled — so downstream consumers can depend on the surface without version drift [VC-C].

Six attack family inputs converge on the governed system under test, producing a chain of hash-linked evidence blocks 600 ATTACK FAMILY 1ATTACK FAMILY 2ATTACK FAMILY 3ATTACK FAMILY 4ATTACK FAMILY 5ATTACK FAMILY 6 GOVERNED DEPLOYMENTunder certification610 620 HASH#1HASH#2HASH#3HASH#4 each entry binds its predecessor
FIG. 1Six attack families (600) exercise the governed deployment (610); each result appends to a hash-linked evidence chain (620) any third party can verify.
Table 1 · Certification evidence vs. typical vendor attestation
PropertyTypical attestationMYTHOS certification
BasisVendor statementExecuted adversarial tests, six families
Evidence formDocument or badgeHash-linked chain, entry-by-entry
Tamper resistanceTrust the authorAny removal or edit breaks the chain
Verifier requirementsChain + sealed 20-name interface only
FreshnessPoint-in-timeContinuous; each run appends
Third-party scenariosRareStanding invitation to evaluators

§ 2 · THE DIRECTIONWhy Regulation Is Moving Toward Continuous Evidence

The NIST AI RMF's MEASURE function frames risk measurement as an ongoing activity rather than a launch-day event [2][3], and the EU AI Act's Article 15 requires accuracy, robustness, and cybersecurity to hold across a high-risk system's lifecycle — not merely at placement on the market [6]. Both point the same direction: assurance as a continuously produced artifact. MYTHOS produces that artifact as a system output, mapped against the adversarial technique landscape that MITRE ATLAS catalogs for AI systems [4].

The invitation to evaluators is standing and deliberate: submit your own attack scenarios for live independent validation. Certification that only its author can run is not certification [VC-C].

§ 3 · IN THE STACKWhat MYTHOS Certifies

Within the portfolio, MYTHOS certifies the governance layer itself: the HCF2-SG trust thresholds and the MRM-CFS execution gates as deployed through SecureAgent ACA Tier C. Its evidence chains also feed the ZGTID consortium as governance telemetry — certified posture, aggregated privacy-preservingly across institutions [VC-C].

The certification boundary is stated as plainly as the capability: MYTHOS certifies the governance layer's resistance at the time of the run, against the families exercised. It does not certify the underlying model's accuracy, does not predict resistance to attack classes outside its six families, and is not an endorsement by MITRE, NIST, or any standards body — those bodies' frameworks are the map its results are plotted against, not the authority issuing them [VC-C].

§ 4 · QUESTIONSFrequently Asked Questions

What does MYTHOS actually certify?

MYTHOS certifies that a specific AI governance deployment resisted six distinct families of adversarial attack at a specific time, and it emits the evidence as a hash-linked chain. It does not certify the underlying AI model’s accuracy, and it is not an endorsement by any standards body — it is executable, re-runnable, third-party-verifiable proof of governance-layer resistance.

Why does hash-linking matter for certification?

Because it removes the need to trust the certifier. Each result entry cryptographically binds to its predecessor, so deleting an inconvenient failure, reordering runs, or editing a result breaks the chain visibly. A verifier needs only the chain and the sealed 20-name interface — not MYTHOS internals and not VectorCertain’s cooperation.

Can external evaluators run their own attacks?

Yes — the invitation is standing. Evaluators and standards bodies can submit their own attack scenarios for live independent validation against a governed deployment. Certification that only its author can execute is attestation wearing a costume; the framework is built for hostile verification.

CONTACTTalk to VectorCertain

Every figure on this page traces to sealed, hash-verified validation artifacts — including findings that contradicted our own published estimates, which we recorded rather than reconciled away. Technical briefings are available for enterprises, evaluators, and standards bodies.

Request the technical briefing

REFERENCES

  1. NIST. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1. nvlpubs.nist.gov
  2. NIST. AI Risk Management Framework (program page). nist.gov/itl/ai-risk-management-framework
  3. MITRE. ATLAS — Adversarial Threat Landscape for Artificial-Intelligence Systems. atlas.mitre.org
  4. European Parliament and Council. (2024). Regulation (EU) 2024/1689 (Artificial Intelligence Act). EUR-Lex. eur-lex.europa.eu
First-party validation artifacts (VectorCertain, sealed and hash-verified):

[VC-C] Patent portfolio and platform engineering baseline — 77-claim hub filing, stack integration claims, 36,181-test regression suite; portfolio documentation, January 2026. Public URLs will replace artifact names when the corresponding research pages publish.

Join the waitlist

Our signup form is temporarily offline while we perform maintenance. Nothing is lost — reach us directly and we'll add you by hand.

PLACEHOLDER · Tally.so form returns here · engineering ticket open

Email us to join