Arthur vs Fiddler AI

Arthur provides an AI Delivery Engine that enables teams to launch, secure, and optimize AI applications across the full lifecycle, from pre-production evaluations to runtime guardrails and production monitoring. The platform supports traditional machine learning, generative AI, and agentic AI with metrics for data drift, accuracy, hallucinations, toxicity, PII detection, and custom evals, while keeping sensitive data local via a federated data-control plane architecture. It targets enterprise AI teams in regulated sectors like banking, healthcare, and insurance, offering deployment flexibility (SaaS, on-prem, AWS, GCP) and integrations like webhooks to Slack/Jira. Distinct from peers, Arthur emphasizes agent discovery & governance and open-source Evals Engine components for real-time performance insights.Arthur homepage, Arthur platform, pricing, PR Newswire Series B, ZoomInfo, LinkedIn

Fiddler AI is an observability, evaluation and guardrails platform that now calls itself "the control plane for enterprise AI agents." It was founded in October 2018 by Krishna Gade (CEO), formerly an engineering manager at Facebook, and Amit Paka (COO), and is based in Palo Alto, California. It has raised $100 million; its latest round was a $30 million Series C led by RPS Ventures in January 2026, with Lightspeed Venture Partners, Lux Capital and Insight Partners among the existing investors taking part. Fiddler says its revenue grew more than fourfold in the 18 months before that round. Customers shown on its site include Nielsen, Integral Ad Science, Qualtrics, Thumbtack, LendingPoint, American Family Insurance and the US Department of Energy. The platform covers two kinds of AI. For predictive models, ML Observability tracks performance, drift and data integrity with root-cause analysis, gives global and local explanations, and runs fairness checks such as disparate impact, equal opportunity and demographic parity on structured, text and image models. For LLM applications and agents, it ingests OpenTelemetry traces (with integrations for LangGraph, LangChain, Strands Agents, Google ADK, Claude Code and LiteLLM), scores them with 80+ built-in metrics, custom evaluators or your own LLM judge, and runs experiments against golden datasets. Its Guardrails, powered by Fiddler's own Centor models, screen prompts and responses for hallucinations (faithfulness), toxicity, jailbreaks, prompt injection and 35+ types of PII and PHI. Fiddler says enforcement takes under 80 ms and runs inside the customer's cloud or VPC with no data leaving it. Deployment options are SaaS, VPC, AWS GovCloud and on-premises. **AI compliance coverage** Fiddler's governance features are audit trails that record every decision, action, evaluation and policy outcome, plus executive dashboards. Its governance page names SR 11-7, HIPAA and NAIC, and lists the EU AI Act among relevant policies; its Control Plane FAQ says its audit trails support the NIST AI RMF, ISO/IEC 42001, HIPAA, SR 11-7, NAIC and the EU AI Act. It doesn't publish control mappings, templates or a model inventory for any of them. Fiddler says it has a SOC 2 Type II report and is HIPAA compliant. In 2023 it released Fiddler Auditor, a free LLM red-teaming and robustness library under the Elastic License 2.0, but its last release was in November 2023 and red-teaming isn't part of the current platform. **Who it's for** Enterprises in financial services, insurance, healthcare and government that run predictive models and LLM agents in production and want monitoring, evaluation and runtime guardrails in one place, often inside their own cloud. It is an observability and enforcement layer, not a governance workflow tool: teams that need AI inventories, risk assessments and policy workflows will need a governance platform alongside it.