# Kapālins platform knowledge base (generated 2026-10-05 from the platform code — do not edit by hand) Kapālins is an Australian-owned enterprise AI security and governance platform. It finds every AI an organisation runs, enforces policy inline, proves each application is acting as it was approved (Autonomous Integrity), and turns compliance into evidence. ## Coverage — five ways in - Proxy (gateway): 15 providers — OpenAI, Anthropic, Google Vertex AI, Google Gemini, Grok (xAI), DeepSeek, Perplexity, Mistral, Cohere, Amazon Bedrock, Hugging Face, Databricks, Azure OpenAI, Qwen, Groq. - Connector (audit logs): 15 sources — Azure Monitor, OpenAI Admin, GitHub Copilot, AWS CloudTrail, Google Workspace, Slack, Zoom, Salesforce, ServiceNow, Glean, Microsoft 365 & Copilot, Snowflake Cortex, Windsurf, Notion, Replit. - Identity: OAuth grants matched against 31 AI apps; SCIM 2.0 from Microsoft Entra ID, Okta, Google, Ping Identity. - Device: a sensor that recognises 69 AI tools and gates prompts, files, local MCP tools and egress. - Probe: scheduled red-team and firewall tests. ## Capabilities 41 shipped capabilities in 12 pillars: - Discover: AI estate inventory, Audit-log connectors, Identity-grant discovery, Device AI discovery, Microsoft AI insights - Protect: AI gateway & prompt firewall, Policy Shield, Gateway self-assurance, Rate & budget fences, Device enforcement, Governed AI Workspace - Detect: Injection & jailbreak, Sensitive-data detection, LLM-judge detection, Session-risk & anomaly, Agentic behaviour monitor, Threat intelligence, Red-team simulation - Agentic & MCP: MCP & A2A gateway, Autonomous Integrity - Supply chain: Model SBOM, Retired-model detection - Govern: Policy generator, Approval queue & playground - Comply: Framework crosswalk, Regulator evidence packs, Incident playbooks & clocks - Score & prove: AI Security Score, AI Trust Index, Risk quantification, Signed Trust Page - FinOps: Cost, chargeback & ROI - Identity: Human & agent identity, SCIM 2.0 provisioning, SSO · MFA · passkeys, Just-in-time access - Integrate: SIEM & ITSM push, Natural-language log search - Operate: Tenants & Guard Units, Partner & MSSP plane, Assessments & board packs ## Plans - Foundation (Illuminate · see your AI): The AI Trust Index, your whole AI estate including shadow AI, peer benchmarking and AI cost. The AI gateway and data security can be added as bridges, so you can try enforcement before moving up. - Growth (Defend · protect your AI): Turn on control: the AI gateway and Policy Shield, red-team probes, threat intelligence and model supply-chain checks, compliance and evidence, policy generation, incident playbooks and budget fences. Deeper capabilities are available as add-ons. - Fortress (Command · prove it and scale): Everything, with no add-ons. Adds regulator evidence packs, SIEM integration, chargeback statements and directory provisioning, plus assurance runs and sealed evidence wherever a feature offers them. - Sovereign (coming soon): Everything in Fortress, deployed on-premises or air-gapped and fully data-resident, for government, defence and tier-1 financial services. Peer benchmarking and the live threat feed need a connected deployment. ## Features - AI Security Score: The security-pillar score across your AI estate, computed from live telemetry. Feeds the AI Trust Index. Always-on. - Peer Benchmarking: Compare your assessment results and posture against anonymised industry peers. Always-on. - Remediation Roadmap: Auto-prioritised, actionable fixes generated from your findings, ranked by their impact on your score. - AI Gateway & Prompt Firewall: The AI gateway plus Policy Shield guardrails — prompt injection, jailbreaks, PII and policy enforcement inline (legacy inference gateway merged in). - AI Estate Discovery: Identify unauthorised and rogue AI deployments across your estate, from every discovery source: the browser extension, your identity provider and the network. - Agentic AI Behaviour Monitor: Real-time monitoring and anomaly detection for autonomous AI agent actions (Copilot, LangChain, CrewAI), against each agent’s own baseline. - Model Threat Simulation / Red-Team: Adversarial probing and red-teaming of connected models (Probe Runner); failures become risk findings and remediation items. - AI Security Threat Intelligence Feed: Real-time threat intelligence specific to AI/LLM vulnerabilities and exploits: feeds, triage, and threat rule-sets enforced at the gateway. - Model Supply-Chain SBOM: Model integrity: provenance, licence and supply-chain bill of materials with SBOM export. - AI Trust Index: The one board number (0–100) — Security, Compliance and Value pillars, every input evidence-linked. - XDR / SIEM Integration: Every enforcement verdict streams to your connected SIEM / ITSM platforms. - Compliance & Evidence: Frameworks, obligations, attestations and the evidence workbench. The regulator evidence pack is its own feature, Regulator Evidence Packs. - AI Incident Response Playbook Engine: Pre-built and customisable incident response playbooks for AI security events. - AI Identity & Access: One identity model, two surfaces: your workforce (single sign-on is on every plan; directory provisioning over SCIM is its own feature, Directory Provisioning) and non-human identities (virtual keys, service identities) with per-key guardrail bindings. - AI Governance Policies: Guided policy generation with statements, apply-to-gateway ACLs, maker-checker four-eyes and drift detection. - AI Cost Observability: Token spend attributed per user, application and non-human identity — the live Value pillar of the AI Trust Index. - Rate-Limit & Budget Control: Per-user / per-app / per-NHI rate limits and token budgets, enforced inline at the gateway. - Chargeback Statements: Chargeback statements: issuing cost statements to your teams and cost centres. Reading cost by team, app and user (showback) is AI Cost Observability, on every plan. - Regulator Evidence Packs: Automated regulator-ready evidence packs — APRA CPS 234, DPDP, EU AI Act — generated continuously from live telemetry. - AI Data Security (DLP): Enterprise data security for AI traffic: jurisdiction-specific PII pattern packs (AU, IN, US, UK, NZ, EU), data-classification ceilings per AI system, custom patterns at scale and redaction evidence reporting. The gateway always keeps the global baseline pack — this feature adds the depth. DLP vendor integrations (Microsoft Purview, Netskope, Forcepoint) on the roadmap. - Agentic AI Security (MCP + A2A): Agentic AI security for tool-calling and agent-to-agent traffic: a governed inline gateway for BOTH MCP (registered tool servers, per-tool allowlists) and A2A (registered peer agents, default-deny method gate, streaming artifact anti-evasion) — Policy Shield on both legs, per-agent identity + tool-namespace binding, call audit trail, and Guard-Unit metering. Agents reach tool servers and peer agents only through the gateway: unregistered = unreachable. - Autonomous Integrity: Everything that acts on your behalf — agents, skills, tool servers, peer agents, credentials and what the sensor finds on devices — measured against the shape you approved, as applications rated by their weakest hop. See reads it; Protect acts on it (accept a shape, allow a surface back, declare and accept a chain); Prove issues chain passports sealed into your evidence chain. - Directory Provisioning (SCIM) & SLA: Automatic user and group provisioning from your identity provider over SCIM 2.0 (Microsoft Entra ID, Okta, Google Workspace, Ping): people and groups are created, updated and deprovisioned as your directory changes, and provisioned groups become policy subjects. Carries the service-level agreement. Single sign-on itself is included on every plan. ## Compliance 29 frameworks, 122+ controls (65 proved by live signals): OWASP LLM Top 10, NIST AI RMF, NIST SSDF, ISO/IEC 42001, ISO/IEC 27001, ACSC Secure AI, Essential Eight, APRA CPS 234, Privacy Act 1988, SOCI Act, ASIC RG 271, ASX Principles, APS responsible AI, MAS TRM · FEAT, PDPA · IMDA, SOC 2, DPDP Act 2023, PIPL, HKMA GenAI, EU AI Act, GDPR, MITRE ATLAS, BNM RMiT, DORA, NIS2, Cyber Resilience Act, APPI, AI Promotion Act, METI AI Guidelines. Available, mapped to your controls at onboarding: Personal Data Protection Law (PDPL) (Saudi Arabia); SAMA Cyber Security Framework (Saudi Arabia); NCA Essential Cybersecurity Controls (Saudi Arabia); AI Ethics Principles (Saudi Arabia); Federal Decree-Law No. 45 of 2021 (PDPL) (United Arab Emirates); DIFC Data Protection Law 2020 and Regulation 10 (United Arab Emirates); ADGM Data Protection Regulations 2021 (United Arab Emirates); Personal Data Privacy Protection Law (Law No. 13 of 2016) (Qatar); Personal Data Protection Law (Law No. 30 of 2018) (Bahrain); Personal Data Protection Law (Royal Decree 6/2022) (Oman). On the roadmap, not yet mapped: AI Basic Act (South Korea); APRA CPS 230 Operational Risk Management (Australia); Personal Data Protection Law (Law No. 27 of 2022) (Indonesia); Colorado AI Act (SB 24-205) (United States). ## Partner programme - Authorised: Entry — regional MSSP, SI practice; Co-branded with Kapālins; 2 sandbox customers - Premier: Established security practice; White-label eligible; client software says “Powered by Kapālins”; 5 sandbox customers - Elite: Carrier, global SI, national MSSP; White-label included; client software says “Powered by Kapālins”; 5 sandbox customers ## Assurance SOC 2 Type II and ISO/IEC 27001 are in progress. Kapālins does not yet hold either certification.