AI Leaders We Cannot Wait to Hear at Ai4 2026

Runtime authorization sits at the center of the enterprise AI conversation, whether the conversation admits it or not.

Every agent that reasons over enterprise data crosses four access points: a prompt input, tool invocations, data retrievals, and response generation. Each of those points can leak, over-return, or bypass the controls a compliance team spent years building.

That’s why PlainID is attending Ai4 2026. The conference gathers the enterprise AI leaders actually shipping models, agents, and RAG pipelines into production. Their sessions surface the exact governance questions the PlainID team works on every day.

Some talks name AI security and trust directly in the title, while others come at it through cost, scale, or business-value framing. All of them create authorization work downstream.

Below, all Ai4 2026 speakers we cannot wait to hear, with each session and the identity angle behind it.

The Ai4 2026 Speakers We Cannot Wait to Hear

Cost conversations about enterprise AI usually skip the authorization angle, and PlainID has watched customers cut cloud spend by tightening agent permissions alone. Our question for Raghu is whether authorization shows up in the FinOps conversation at Barclays or stays in a different room.

This is the one question every CISO we talk to is trying to answer for their board. Biswa’s framework at JPMorgan should validate or challenge the criteria PlainID prospects surface with us every week.

Bank reference architectures always name the data, model, and orchestration layers, but only some name the runtime authorization layer that sits underneath. Our eye is on Bhairav’s diagram to see whether the enforcement plane makes the reference architecture or lands as a footnote.

CVS runs one of the most complex healthcare AI programs in the US, spanning payer, provider, and retail pharmacy. Our design question for Tony is how his team built consent and purpose-based access controls that work under HIPAA and state-level rules without breaking the member experience.

Mae-Beth is one of the very few speakers with both AI governance and privacy in the same job title. Aerospace-scale governance means export controls, model provenance, and cross-border data flow all inside one policy stack. Whatever the session covers, PlainID is in the front row.

Broadridge processes trades for most of the global buy side, and Partha’s team defends that infrastructure. The angle we care about is which controls hold when the attacker uses the same generative tools you do, and where identity-based authorization slots into that defense.

Fanny’s session maps most cleanly to what PlainID actually builds. The comparison we want is how Roche layers guardrails, enforcement, and observability into a single control plane, and whether pharma treats the pattern as platform work or as governance work.

L’Oréal runs consumer AI across dozens of regulated markets, which means GDPR, LGPD, PIPL, and emerging AI rules all sit on the same pipeline. Our question for Saad is how L’Oréal handles the per-market authorization overlay without cloning the whole stack for every jurisdiction.

Erik created the x402 protocol, so this fireside puts non-human identity at its most extreme: agents that hold wallets and spend money. Two things we’re listening for are how Coinbase authorizes an agent that can transact, and where accountability lives when the agent makes the wrong call.

RAG is where AI meets your most sensitive data, and at Morgan Stanley that means client records, positions, and communications under fiduciary rules. What we want to hear is how Nicholas’s team filters retrieval by role, purpose, and jurisdiction before it reaches the model, and where the audit trail sits.

The shift from recommendation systems to autonomous decision engines on industrial floors changes the identity model entirely. Caleb’s session should tell us how Siemens decides which agents get to act on which equipment, and how the audit record survives when something goes wrong.

Airbus’s partnership model spans OEMs, suppliers, airlines, and regulators, which makes cross-organization identity federation the real constraint on scaling AI from pilot to production. Reena’s session should tell us whether the sticking points map to technology, culture, or the reporting split between AI and identity teams.

COCO is one of the few named production agentic systems disclosed publicly by a top-five pharma. Every field it touches sits under consent, contract, or compliance rules. Two questions we want answered are how Pfizer scoped which agent gets which permissions, and how the guardrails hold up in production.

AI-assisted code shipping to millions of consumer-banking devices is the highest-stakes DevSecOps use case in production. Where identity and permission checks fit inside the defense layers is what we came for, especially when the AI writing the code does not know which permissions are real.

Most unusual session on the roster. Every innovation Cigna ships eventually crosses HIPAA and provider-data boundaries. Our plan is to soak up Dan’s framing on where new ideas come from, then apply the governance lens after the fact.

Capital One holds a public track record at the intersection of DevSecOps and AI. Ravi’s session should show us how the security review process wires into AI-generated pull requests, and where identity and permission checks land inside the pipeline.

Deutsche Bank operates under DORA, which turns scaling compute for AI without scaling the control plane around it into a regulatory question. For Frank, the value is in how his team maps infrastructure decisions to the operational-resilience framework the regulator is actually watching.

Gal will cover runtime controls across the full agentic AI flow, guardrails at every access point, and central authorization for every human and machine identity.

The Identity Gap Behind Every Ai4 2026 Session

Notice what almost none of the session titles include. Identity. Authorization. Access control. Runtime enforcement.

We don’t think this pattern is an accident. Enterprise AI leaders ship models, agents, and business outcomes for a living. Identity and authorization sit in a different reporting line, usually under the CISO or Chief Identity Officer. And that reporting split is where audit-committee questions land.

Every AI system the speakers describe on stage passes through four control points:

• The prompt input
• The tool or MCP invocation
• The data retrieval
• The response generation

PlainID enforces policy at all four points, in real time, using one PBAC policy language across apps, APIs, data, microservices, and AI agents. Gartner names PlainID in three 2026 reports covering this exact space, including the Market Guide for Guardian Agents, Reference Architecture Brief for IAM for AI Agents, and Sector Overview: Information Security and Risk Management, 2Q26. Our Fortune 500 clients like Wells Fargo, Cisco, Accenture, and FM Global run this pattern in production. The audit committee gets an answer their regulator accepts.

Meet PlainID at Ai4 2026

Ai4 2026 runs August 4 to 6 at The Venetian in Las Vegas. The speakers above will spend three days explaining what production enterprise AI actually looks like. PlainID will be on site to help identity, security, and IAM leaders translate that work into policy their audit committees can defend.

Find out booth #1045 or book a 30-minute session with the PlainID team on-site. Bring a specific agent use case, a compliance question, or an audit gap you are trying to close. We will walk through how runtime authorization applies, using your stack, your data platforms, and your regulator.