Available nowADVERSARIAL TEST-SET REGISTRY

ATR

Govern provenance, licensing, applicability and usage of adversarial items, and turn confirmed vulnerabilities into mandatory regression assets for future releases.

Product flow

01Verified registry
02Generalization & regression
03Human verdict & report
From input to a verifiable outcome

DESIGNED FOR

Organizations that need to manage prompt injection, hijacking and jailbreak risk as a continuous defensive regression program rather than a one-off test

OUTCOMES

Start with customer change, not a feature list

Verified live-site capabilities and current product sources are reframed around the customer workflow and operating outcome.

01

Stop rebuilding test sets

Accumulate adversarial tests that were previously scattered by product into a registry with common schema, taxonomy and provenance.

02

Separate discovery from recurrence

Run new-sample generalization and mandatory retesting of confirmed vulnerabilities under distinct rules.

03

Keep attack content contained

Restrict access to operational attack prompts and expose only approved aggregates to general users and partners.

CAPABILITY SYSTEM

The capabilities that make the product work

These are product-level capabilities customers can adopt and operate—not isolated buttons or controls.

01

Item registry

Manage item IDs, taxonomy, dependencies, provenance, license and lifecycle state together.

02

Applicability

Select only items applicable to country, industry, company, service, brand, product, revision and evaluation purpose.

03

Generalization track

Sample deterministically from the unused pool and prevent reuse for the same product and purpose.

04

Regression track

Promote human-confirmed vulnerabilities into product regression suites and force their inclusion in future releases.

05

Campaign & manifest

Fix item, product, model, judging criteria and execution revisions in a manifest for reproducibility.

06

Human verdict & reporting

Use automated judging as input, require human confirmation for promotion and produce aggregate-safe reports.

OPERATING FLOW

How it works after adoption

  1. 01

    Intake, register and normalize

    Register human red-team, approved external and generated items under one contract.

  2. 02

    Hygiene, rights and applicability

    Validate duplicates, licenses, permitted use and target applicability.

  3. 03

    Run generalization or regression

    Choose the appropriate track and execute the campaign with a fixed manifest.

  4. 04

    Verdict, promotion and report

    Use human verdicts to update consumed or regression state atomically and publish aggregate results.

TRUST & DEPLOYMENT

The operating environment and responsibility boundary are part of the product.

01

On-premises operations portal

Provide a read-only portal for QA and service experts to review data rights, lifecycle and campaign status.

02

Raw-content isolation

Keep attack and response payloads out of general web projections and enforce purpose limitation and access control.

03

Product-specific adapters

Validate chatbot, agent and copilot defenses through adapters bound to approved endpoints and revisions.

VERIFIED SCOPE

Evidence and boundaries behind the public copy

  • A two-track contract separating generalization consumption from mandatory regression replay
  • Item-level provenance, licensing, permitted use and export controls
  • Regression promotion requires a human-confirmed verdict, not an automated judge alone