Uploaded June 2026 | Updated September 2026, 2 weeks ago
What happens when an AI hallucination isn’t just embarrassing—but creates regulatory, legal, and financial risk?
In this session, BlackRock shares how it approaches responsible AI at enterprise scale, where AI systems support investment professionals managing trillions of dollars in assets. Through real-world examples and lessons learned from deploying AI in highly regulated environments, the talk explores why trust, governance, and risk management must be treated as core product requirements rather than afterthoughts.
The session walks through BlackRock’s responsible AI platform, including real-time guardrails, evaluation pipelines, observability systems, compliance workflows, and governance processes designed to reduce risk while accelerating AI adoption. Learn how BlackRock built a framework for deploying AI safely in financial services and why responsible AI must become the foundation of every production AI system operating in regulated industries.
Timestamps
00:00 Why AI Hallucinations Matter in Financial Services
01:17 The Cost of Getting AI Wrong in Regulated Industries
02:15 Building Responsible AI at BlackRock
03:20 Why Responsible AI Must Be a Product, Not a Feature
05:15 Lessons from Early Agent Platform Deployments
06:18 Catching a Critical AI Hallucination Before Production
07:42 Rebuilding the Responsible AI Roadmap
08:08 Shipping Trust Systems, Not AI Features
09:00 The BlackRock Responsible AI Platform
11:05 Real-Time AI Guardrails and Moderation Systems
12:42 Custom Guardrails for Financial Services
13:06 Building an Investment Advice Guardrail
14:38 Working with Legal and Compliance Teams
15:15 Offline and Online Evaluation Pipelines
16:35 Human Review and Production Evaluations
17:20 Measuring Success Through Confidence and Trust
18:18 Velocity to Confidence: BlackRock’s Key Metric
19:18 The Human Element of Responsible AI
Key Takeaways
• In highly regulated industries, AI failures can create legal, compliance, and reputational risks that far exceed typical consumer-facing AI mistakes.
• Responsible AI cannot be treated as a final checkpoint or compliance layer—it must be designed as a core product from the beginning.
• Effective AI governance requires real-time guardrails, continuous evaluation, observability, and close collaboration between engineering, legal, and compliance teams.
• Industry-specific guardrails are essential for production AI systems operating in regulated environments such as financial services.
• Trust remains the most important metric for enterprise AI adoption, making responsible AI a business requirement rather than a technical feature.
--
#ResponsibleAI #BlackRock #EnterpriseAI #AIGovernance #FinancialServices #AICompliance #AIObservability #LLMOps #AIAgents #ArizeObserve
🔗 Try Arize AX & Phoenix OSS: arize.com
🔔 Subscribe for weekly content on LLMs, agents, and evaluation: youtube.com/@arizeai?sub_confirmation=1
What happens when an AI hallucination isn’t just embarrassing—but creates regulatory, legal, and financial risk?
In this session, BlackRock shares how it approaches responsible AI at enterprise scale, where AI systems support investment professionals managing trillions of dollars in assets. Through real-world examples and lessons learned from deploying AI in highly regulated environments, the talk explores why trust, governance, and risk management must be treated as core product requirements rather than afterthoughts.
The session walks through BlackRock’s responsible AI platform, including real-time guardrails, evaluation pipelines, observability systems, compliance workflows, and governance processes designed to reduce risk while accelerating AI adoption. Learn how BlackRock built a framework for deploying AI safely in financial services and why responsible AI must become the foundation of every production AI system operating in regulated industries.
Timestamps
00:00 Why AI Hallucinations Matter in Financial Services
01:17 The Cost of Getting AI Wrong in Regulated Industries
02:15 Building Responsible AI at BlackRock
03:20 Why Responsible AI Must Be a Product, Not a Feature
05:15 Lessons from Early Agent Platform Deployments
06:18 Catching a Critical AI Hallucination Before Production
07:42 Rebuilding the Responsible AI Roadmap
08:08 Shipping Trust Systems, Not AI Features
09:00 The BlackRock Responsible AI Platform
11:05 Real-Time AI Guardrails and Moderation Systems
12:42 Custom Guardrails for Financial Services
13:06 Building an Investment Advice Guardrail
14:38 Working with Legal and Compliance Teams
15:15 Offline and Online Evaluation Pipelines
16:35 Human Review and Production Evaluations
17:20 Measuring Success Through Confidence and Trust
18:18 Velocity to Confidence: BlackRock’s Key Metric
19:18 The Human Element of Responsible AI
Key Takeaways
• In highly regulated industries, AI failures can create legal, compliance, and reputational risks that far exceed typical consumer-facing AI mistakes.
• Responsible AI cannot be treated as a final checkpoint or compliance layer—it must be designed as a core product from the beginning.
• Effective AI governance requires real-time guardrails, continuous evaluation, observability, and close collaboration between engineering, legal, and compliance teams.
• Industry-specific guardrails are essential for production AI systems operating in regulated environments such as financial services.
• Trust remains the most important metric for enterprise AI adoption, making responsible AI a business requirement rather than a technical feature.
--
#ResponsibleAI #BlackRock #EnterpriseAI #AIGovernance #FinancialServices #AICompliance #AIObservability #LLMOps #AIAgents #ArizeObserve
🔗 Try Arize AX & Phoenix OSS: arize.com
🔔 Subscribe for weekly content on LLMs, agents, and evaluation: youtube.com/@arizeai?sub_confirmation=1
