The Road to Accountable AI

The episode is densely packed with concrete technical concepts and actionable governance frameworks for agentic AI. Venkat articulates novel distinctions (model safety vs. agent safety, measured autonomy, semantic containment, trust boundaries, autonomy budgets) that most operators haven't deeply considered. However, some sections involve restating concepts or throat-clearing rather than pure insight delivery.

“Traditional controls were largely designed for fixed workflows, known permissions, and predictable software behavior. They were not designed for software that can decide at runtime what tools to call or what context to use or what actions to take.”

“The control plane needs to shift to where the decisions are made versus being a static set of steps”

Venkat presents fresh frameworks around the execution boundary as the critical governance layer, and introduces specific original thinking on measured autonomy, identity for agents (DIDs), and fleet-level policies. However, some concepts (least privilege, audit trails, prompt injection layering) are well-established security practice, and the discussion occasionally relies on rehashing standard AI governance talking points.

“we focus on is how do you govern behavior at the execution boundary, right? Like this is exactly where you have prompts, you have memory, you have identity, you have tools, and you have policies”

“our view is every enterprise agent needs a distinct, uh, machine-readable identity... DID is decentralized identifiers... it's not just a display name like a HR assistant or a, a finance bot, but a verifiable identity that can be resolved- Checked and tied to cryptographic keys”

Venkat is a credible practitioner with 20+ years building platform products for enterprise technology adoption, direct experience at Rivian (automotive, high-stakes domain), and is actively building governance infrastructure for agents at Comply. He speaks from operational experience rather than theory, though he is also a startup founder/pitch guest and not a large-scale current operator.

“I have spent over 20 years building platform products that help enterprises adopt new technology safely and at scale. What I saw at Rivian, and what I am frankly seeing across the market, is that agents are becoming incredibly easy to build.”

“Venkat Siva, co-founder and CEO of Comply”

The episode includes concrete domain examples (automotive service triage agents, customer support agents, procurement/AP payment workflows, OpenClaw shadow agents, forklift hallucination risks) and specific technical references (DIDs, MCP, OWASP's lethal trifecta, COSAI paper identifying 40+ threats in MCP). However, it lacks named customer case studies, specific breach examples, quantified metrics, and dollar figures that would elevate specificity further.

“let's take an agent that looks for emails or basically looks for a trigger from a customer to say, 'Show me... Can you guys get me a demo for this product?'... in a least privileged scenario... you basically say, this system or a process, processes gets the exact permission that it needs”

“COSAI, which is, uh, a community we are part of, um- Mm-hmm ... uh, just basically wrote a paper about sort of how many threats are available just in the base MCP specification. Mm-hmm. There are over 40 threats across 12 categories”

Kevin asks good foundational questions and occasionally pushes back (e.g., "why don't agents just belong as a feature under existing MLOps systems?"), but the conversation is largely linear and affirming. Kevin rarely challenges Venkat's claims, doesn't demand metrics, and mostly accepts his framing. The follow-ups are clarifying rather than investigative. There's minimal productive disagreement or pressure-testing of assumptions.

“Let me just push you a little bit more on, on that. There are plenty of tools now for MLOps and, uh, AI security. Why is, uh, why, why don't agents just belong as a feature under those platforms?”

“Can you give an example of a situation where, where the model was safe but still some harm came about from the agent?”