AI Guardrails for Enterprise Deployments
summary
Atomicwork runs an AI-native ITSM that masks PII, shows an “generated by AI” disclaimer, and requires agent approval before replacing user content; a cited Microsoft stat says 90% of power users use generative AI daily and the team reports a 90% reduction in GPT-4 token cost over six months. Nine guardrails focus on reliability, resilience and responsibility: use Microsoft Procedo for PII detection, prefer licensed models like Azure OpenAI or Bedrock (or domain models such as Gorilla LLM), run continuous prompt-injection tests and SiteLabs pen testing, and log provenance from SharePoint/Notion.
AI-written summary · about
chapters (6)
- 0:00Introduction to AI guardrails
- 1:37Enterprise AI adoption trends
- 4:08Nine AI guardrails for enterprises
- 8:47Resilient AI system design
- 10:34Prompt and model security
- 13:00Output guardrails and monitoring
related talks
45mSecuring and Governing Generative AI: Learnings from Microsoft
47mData Security Considerations for AI Adoption
24mAI Governance for developers: A practical framework for deploying autonomous coding agents safely…
26mSponsored Session: Navigating Open Source AI with Guardrails
42mSecure your AI application transformation with Security
28mAI for Everything After Code
20mAI Red Teaming Agent: Azure AI Foundry
34m“You Can’t Just Turn It On”: The Hidden Work Behind Enterprise AI