Guest: Randy Bias (LinkedIn)
Company: Mirantis
Show Name: An Eye on AI
Topic: AI Governance

As enterprises scale their use of AI agents, the challenge isn’t just performance or cost—it’s control. Security and compliance frameworks are still catching up to the reality of agentic AI, and the rules differ wildly across borders. Randy Bias, VP of Strategy & Technology at Mirantis, says that’s why global organizations need an adaptable framework for governance.

Beyond Compliance: The Rise of the “Virtual Air Gap”

Bias explains that data protection in AI-driven systems now requires new architectural thinking. “You want to make sure certain kinds of data don’t leave your four walls,” he says. For highly sensitive information—like healthcare or financial records—enterprises need a virtual air gap, where AI agents interact only with local inference engines instead of external LLMs. Less sensitive workflows can still connect to public frontier models, but within strict boundaries.

This layered approach ensures that enterprises can harness AI’s full power without exposing data to unauthorized systems—a key capability built into Mirantis’ MCP AdaptiveOps offering.

Customizing Guardrails for a Global World

But security is just the start. Bias highlights how AI compliance now intersects with culture and policy. “If you’re in Thailand,” he notes, “it’s against the law to say anything against the King.” A single global AI policy won’t account for nuances like that. AdaptiveOps allows organizations to define region-specific guardrails, embedding rules that reflect local laws, corporate ethics, and language sensitivities.

These guardrails govern every layer of AI interaction—from agent-to-agent communication to data exchange with internal or external models. They help enterprises maintain compliance not only with regulations like GDPR or HIPAA, but also with cultural expectations and internal governance policies.

Learning to Govern Non-Deterministic Systems

Perhaps the biggest shift, Bias observes, is the mindset. Traditional software followed predictable logic; AI doesn’t. “In the past, you didn’t have to worry about arbitrary things happening—but now we do because AI is non-deterministic,” he says. That uncertainty demands flexible, evolving guardrails rather than static policies.

Mirantis’ AdaptiveOps takes that reality head-on by combining technical controls with continuous governance support. It’s not just about deploying AI securely—it’s about operating responsibly in a world where the rules can change overnight.