Add selected Workbench capabilities through bounded OEM and partner integrations
AI SECURITY LABS
Labs exposes selected prototypes, reference implementations, scenarios, fixtures, and public-safe datasets used to explore AI security methods. A Labs item is not automatically a supported Workbench product, production integration, or commercially available capability.
6
Published scenario rollups
Aggregate rows in the current public scenario snapshot.
15
Published attack packs
Attack-pack records in the current public range snapshot.
8
Maintained tool adapters
Adapters represented in the current public range registry.
15
Mapped ATLAS techniques
Distinct MITRE ATLAS technique mappings in the current public scenario registry.
Graduated Workbench capabilities
Some capabilities that originated as experimental or research surfaces now have canonical Workbench identities. Labs does not maintain competing product definitions for them.
Public website trust intelligence for AI claims, legal clarity, security trust, governance evidence, and cross-page consistency.
Explore Trust Scanner
Scenario execution, adversarial test coverage, and evidence capture under the canonical Workbench capability.
Explore Adversarial Range
Lab modules
Status describes the public Lab record. Prototype and experimental utility labels are boundaries, not implied support or commercial availability.
reference implementation
Unified framework navigation across OWASP LLM Top 10, NIST AI RMF, MITRE ATLAS, and ISO 42001 — with directional cross-framework mappings, evidence prompts, and scorecard bridges.
Intended use
Explore maintained directional framework cross-references and evidence prompts.
reference implementation
Launch into the public AI Security Scorecard and the organizational AIPSA scorecard when you need the maturity model.
Intended use
Review the public scorecard structure and directional baseline model.
experimental utility
Deterministic prompt policy review across 15 rules: injection resistance, identity anchoring, output handling, secret detection, and RAG context isolation. Includes KB/corpus scanner.
Intended use
Demonstrate deterministic prompt-policy checks; not a supported peer Workbench product.
experimental utility
Paste a RAG pipeline JSON config and get instant findings across authorization gaps, tenant isolation failures, over-retrieval, document provenance, and sensitive context exposure.
Intended use
Demonstrate retrieval configuration analysis and fixture review.
Related Workbench capability
experimental utility
Analyze agent tool configurations for overly broad scopes, missing human-approval gates, unsafe side effects, ambiguous tool descriptions, and privilege escalation paths.
Intended use
Demonstrate agent permission and approval analysis without creating a peer product.
Related Workbench capabilities
experimental utility
Paste model output, select the sink type (HTML, Markdown, JSON, tool call, email, DB, code), and get deterministic safety analysis across injection, leakage, and side-effect risks.
Intended use
Demonstrate deterministic output-sink checks against public fixtures.
fixture
A structured library of 12 attack probes across 10 categories. Record blocked/detected/degraded/passed outcomes per probe and export a full evidence session as JSON or Markdown.
Intended use
Exercise structured prompt-injection probes and review export schemas.
scenario library
Review model artifacts, dependency paths, provenance gaps, and poisoning scenarios. Identify integrity weaknesses using existing attack packs without rebuilding the artifact scanner.
Intended use
Practice model-integrity and supply-chain analysis against public-safe fixtures.
scenario library
Analyze poisoned persistent memory, session history, and context influence in agentic systems. Trace how contaminated entries change downstream decisions.
Intended use
Study persistent-memory and context-poisoning scenarios.
scenario library
Review tenant data, prompt traces, PII fixtures, and retrieval behavior to identify AI-specific data exposure failures. Interactive evidence builder with localStorage persistence.
Intended use
Practice cross-tenant and sensitive-context evidence review with synthetic fixtures.
scenario library
Study hidden instructions in images, SVGs, steganographic content, and document-derived context. Design test plans for vision-enabled AI systems.
Intended use
Explore multimodal instruction and content-boundary test design.
reference implementation
Convert AI policy claims, controls, and crawler signals into governance evidence. Build a control evidence matrix with public-safe language.
Intended use
Practice careful claim-to-control evidence mapping with public-safe fixtures.
reference implementation
Practice discovering AI providers, model dependencies, data flows, and shadow AI surfaces. Create an AI system inventory artifact for governance and product security.
Intended use
Practice system-boundary and AI inventory artifact construction.
reference implementation
Use the existing Threat Canvas to produce an AI trust-boundary threat model. Place boundaries, find abuse paths, and assign controls for engineering review.
Intended use
Practice AI trust-boundary threat modeling against a reference scenario.
Related Workbench capability
scenario library
Review AI trace, prompt, completion, retrieval, and tool-use logs. Identify telemetry gaps and produce a forensic evidence chain for AI abuse scenarios.
Intended use
Review synthetic AI telemetry and construct a forensic evidence chain.
Related Workbench capability
scenario library
Walk through an AI abuse incident, classify the event, preserve evidence, decide escalation, and produce an after-action plan using existing incident drill scenarios.
Intended use
Practice evidence preservation, escalation, and after-action decisions in a synthetic incident.
Reference datasets and fixtures
These assets support reproducibility, demonstrations, schema review, and method development. Their presence does not establish production deployment, customer adoption, connector support, or independent validation.
Versioned product catalog for the public labs shell.
Curated scenario rows and control-evidence rollups.
Status, scope, caveat, run window, and data endpoints.
Public-safe scorecards and range metrics.
Monthly synthetic-media and prompt-abuse tracking.
Public-safe AI Trust and Governance scorecard sample with required caveat.
They show that a public schema, fixture, scenario, or demonstration exists at the stated version and scope. They do not prove production use, customer results, commercial support, external acceptance, connector compatibility, or coverage beyond the published record.
This product shell is curated from public-safe outputs and native in-repo routes.