Arthur

Arthur is an enterprise AI observability, security, and governance platform that helps teams ship reliable AI agents and models, with continuous monitoring, guardrails, and a first-of-its-kind Agent Discovery & Governance (ADG) layer. It is built for security, risk, and AI teams that need to see, evaluate, and control agentic AI in production.

Arthur (Arthur AI) is a New York-based AI reliability and governance company founded in 2018 by Adam Wenchel and co-founders. Its platform provides a single control plane spanning three layers of the AI stack: observability (continuous monitoring of deployed models and agents), security (pre- and post-LLM guardrails that block sensitive-data leaks, prompt injection, hallucinations, and harmful content), and governance (evaluation, policy enforcement, and audit). In December 2025 Arthur launched its Agent Discovery & Governance (ADG) platform, which it positions as the industry's first product purpose-built for the agentic era rather than retrofitted from classic ML monitoring; ADG discovers 'shadow agents' via OpenTelemetry streams, MCP server monitoring, network traffic, and platform APIs (Vertex AI, AWS Bedrock, Azure AI Foundry), then applies runtime enforcement and continuous unsupervised evaluations for hallucination risk, response completeness, and goal accuracy. Arthur runs on a federated architecture that keeps sensitive inference data inside customer VPCs, offers SaaS, on-premises, and cloud-marketplace (GCP/AWS) deployment, and maps controls to the EU AI Act, NIST AI RMF, and ISO 42001. The company has raised roughly $63 million, including a $42 million Series B in 2022 led by Acrew Capital and Greycroft.

Category

Governance & Security

Pricing

Contact for pricing

Target Market

CISOs, CIOs, AI/ML Governance Leaders, Data Scientists

Founded

2018

Headquarters

New York, New York, United States

• ✓
  Agent Discovery & Governance (ADG)

  Detects agents across the enterprise via OpenTelemetry, MCP monitoring, network traffic, and platform APIs, surfacing shadow agents deployed without central IT awareness.

• ✓
  Runtime guardrails

  Pre-LLM checks screen for sensitive data and injection attacks while post-LLM validation catches hallucinations and harmful outputs, with self-correcting loops.

• ✓
  Continuous evaluation

  Unsupervised binary evaluations run on production traffic to measure hallucination risk, response completeness, goal accuracy, and topic consistency.

• ✓
  Observability on open standards

  Built on OpenInference/OpenTelemetry conventions for end-to-end visibility across LLM calls, RAG retrievals, and tool invocations.

• ✓
  Federated, in-VPC deployment

  A federated architecture keeps sensitive inference data inside the customer's VPC, with SaaS, on-prem, and GCP/AWS marketplace options.

• •
  Shadow-agent discovery

  Find and inventory autonomous agents running across multi-cloud environments before they create compliance or security risk.

• •
  LLM output safety

  Block prompt-injection, PII leakage, and hallucinated or harmful responses in real time before users see them.

• •
  AI audit and compliance

  Generate evaluation and audit evidence mapped to the EU AI Act, NIST AI RMF, and ISO 42001.

Enterprise pricing is not publicly disclosed; Arthur also maintains open-source components and is available on the Google Cloud and AWS marketplaces.

This page was written from 3 sources, 1 on domains other than arthur.ai.

1. 1.arthur.ai — arthur.aivendor
2. 2.arthur.ai — best ai governance platforms 2026vendor
3. 3.prnewswire.com — arthur raises 42m in series b funding as ai adoption soars 3