What governance practices help organizations adopt AI agents responsibly?

Most answers emphasize implementing guardrails, controls, and monitoring, while others highlight case-by-case governance, least-privilege access, and observability as key responsible AI practices.

10 answers

Answers

  • Dana Simberkoff

    Chief Risk, Privacy and Information Security Officer · AvePoint

    Case-by-case governance

    Dana Simberkoff says her organization views AI use cases on a case-by-case basis, with a governance process for approving future use cases once policy is in place, because no two things fit the same stack.

    AvePoint · Watch on YouTube
  • Ekta Chopra

    Chief Technology & AI Officer · E.L.F. BEAUTY

    Risk governance

    Ekta Chopra recommends that enterprises start with policies and risk governance, along with an observability strategy and an inventory of systems agents can access, because AI agents can learn, behave on their own, or hallucinate and cause damage.

    The answer starts at 42:51.

    CEOs and ABCs · Listen on Apple Podcasts
  • James J Azar

    Chief Information Security Officer · Artera Services

    Least-privilege access

    James J Azar says the guidance emphasizes least-privilege access, with agents receiving only the permissions necessary to perform their defined functions.

    CISO Talk by James Azar · CISO Talk by James Azar · Read
  • Tyler Lynch

    Outbound CTO - IBM Automation Portfolio · Ibm

    Sandboxing agents

    Tyler Lynch says he and a colleague had the idea that agents making bad financial behavior could be sandboxed and isolated, with policies possibly revoking an agent's API key until a human reviews it.

    SiliconANGLE theCUBE · Watch on YouTube
  • Rama Akkiraju

    VP, Enterprise AI · Nvidia

    Observability

    Rama Akkiraju says observability is key, especially for understanding agent behavior before an agent can be certified.

    Imagination in Action · Watch on YouTube
  • Mihai Criveti

    Distinguished Engineer, Agentic AI · Ibm

    Guardrails

    Mihai Criveti says he has been calling for guardrails to be implemented for large language models, for agents talking to other agents, and for MCP servers, citing the risk that models can call thousands of agents to scale an attack.

    The answer starts at 5:57.

    Mixture of Experts · Listen on Apple Podcasts
  • Rashmi Shetty

    Senior Director - Enterprise Generative AI Platform · Capital One

    Guardrails

    Rashmi Shetty says organizations adopting agentic AI need guardrails, which she describes as new safety and governance mechanisms introduced into agentic AI systems.

    The answer starts at 3:07.

    Computer Weekly Downtime Upload · Listen on Apple Podcasts
  • Paula Goldman

    Chief Ethical and Humane Use Officer & EVP, Product · Salesforce

    Establish controls and testing frameworks

    Paula Goldman says responsible adoption is about setting up the right framework, with controls inside the technology to oversee it and get quality and accuracy assurance, and testing to see whether it gives the right results.

    The answer starts at 5:12.

    Thinkers & Ideas · Listen on Apple Podcasts
  • Rob Allen

    Chief Product Officer · Threatlocker

    Identify usage, enforce guardrails

    Rob Allen says organizations should start by identifying where these tools are used and then figure out what access they need, arguing that unrestricted access is not something they need and that tools can operate within guardrails.

    Security Weekly - A CRA Resource · Watch on YouTube
  • Jacob Thomas

    Manager of Information Security · Texas Children'S Hospital

    Guardrails, monitoring, auditability

    Jacob Thomas says his approach relies on mitigating and compensating controls, and that planning for them correctly can reduce the attack surface and limit what threat actors can do, an approach he says is not unique to AI.

    SiliconANGLE theCUBE · Watch on YouTube