For more than a century, the modern corporation has been built around a relatively stable assumption: people make decisions, people execute transactions, systems record what happened, and auditors periodically examine the evidence.
That assumption is beginning to break.
AI is moving from answering questions to taking actions. Agents can retrieve information, make recommendations, initiate workflows, communicate with customers and suppliers, update systems of record, generate financial analysis, execute transactions, and increasingly coordinate with other agents. The enterprise is becoming agentic.
The implications extend far beyond technology.
When software can act on behalf of a company, the fundamental questions of corporate governance begin to change. Who gave the agent authority? What was it permitted to do? What information could it access? What controls governed its actions? What happened when circumstances fell outside those controls? Who remains accountable? And, perhaps most importantly for boards and audit committees, how do we know?
These questions are why the Arthur and Toni Rembe Rock Center for Corporate Governance at Stanford Law School and the Alpha Institute for AI Governance are launching the Stanford Directors’ College Agentic Audit Forum, taking place October 23–25, 2026 at Stanford Law School.
The invitation-only, three-day program will bring together directors, CFOs, controllers, chief audit executives, general counsel, corporate secretaries, risk and compliance leaders, academics, and practitioners to examine how governance, controls, assurance, and accountability must evolve when AI systems can act on behalf of the enterprise.
The subject is timely because we are entering a period in which the speed of enterprise activity may increasingly exceed the speed of traditional governance.
From the AI Enterprise to the Agentic Enterprise
The first phase of enterprise AI was largely about intelligence.
Companies used machine learning to identify patterns, improve forecasts, personalize experiences, automate routine analysis, and eventually generate text, images, code, and other content. Generative AI dramatically expanded what machines could produce, but in most environments a human remained responsible for deciding what happened next.
Agentic AI changes that relationship.
An agent does not simply provide an answer. It can pursue an objective. It can use tools, access systems, invoke APIs, make intermediate decisions, communicate with other agents, and take actions within whatever authority the enterprise has granted it.
That distinction may sound technical. From a governance perspective, it is profound.
An AI system that recommends a payment is one thing. An agent authorized to initiate the payment is another.
A model that drafts a supplier agreement is one thing. An agent capable of negotiating terms, routing approvals, updating procurement systems, and triggering payment obligations is another.
A system that identifies an accounting anomaly is one thing. An agent capable of making entries, reconciling accounts, generating evidence, and preparing financial reporting is another.
Once AI moves from information to action, governance must move with it.
We are no longer governing only models. We are governing authority.
The Agentic Economy Will Operate at Machine Scale
The same transformation taking place inside companies will increasingly take place between them.
Agents will interact with other agents representing customers, suppliers, financial institutions, marketplaces, insurers, professional-services firms, and eventually regulators. They will request information, evaluate counterparties, negotiate terms, make purchases, execute workflows, and continuously evaluate risk.
This is the emerging agentic economy.
Its defining characteristic will not simply be greater automation. It will be speed and scale.
Human organizations operate primarily at human cadence. Meetings happen weekly or monthly. Committees convene quarterly. Policies are reviewed annually. Audits examine defined periods. Exceptions move through approval chains designed around the time required for people to evaluate them.
Agents operate differently.
They can execute thousands of actions while a committee meeting is still being scheduled.
That creates what may become one of the defining corporate governance challenges of the next decade: the governance speed gap.
If operations take place continuously, but oversight remains episodic, the distance between action and accountability grows.
The solution cannot simply be more meetings, more policies, or larger compliance teams. Governance itself must become more continuous, more machine-readable, and increasingly capable of operating at the same speed as the systems being governed.
Audit sits at the center of this transition.
The Audit Trail Becomes the Action Trail
The audit profession has always depended on evidence.
What happened?
Who authorized it?
What controls applied?
Was the transaction properly recorded?
Can the evidence be independently verified?
In an agentic environment, these questions remain essential, but the nature of the evidence changes.
An audit trail can no longer consist only of records documenting the final outcome. Organizations will increasingly need to understand the chain of authority and behavior that produced it.
Which agent acted?
Under whose authority?
For what purpose?
Using which model?
With access to what data?
Which tools did it invoke?
What policies applied?
What limits were established?
Did another agent participate?
Was human approval required?
Was an exception triggered?
What evidence demonstrates that the control actually operated?
This is the beginning of behavioral assurance.
Rather than examining only what a system was designed to do, enterprises will need evidence of what agents actually did.
That evidence may eventually become as important to corporate governance as financial records are today.
The Forum will examine precisely these emerging issues, including agent authority and permissions, internal control over financial reporting in agentic workflows, AI-generated evidence, external auditor use of AI, AI disclosure and financial reporting, agent identity, cybersecurity, third-party risk, and continuous monitoring and assurance.
Internal Control Has to Follow the Work
The implications for internal control over financial reporting are particularly important.
Much of today's control environment assumes identifiable individuals occupying defined roles within established workflows.
Someone prepares.
Someone reviews.
Someone approves.
Someone records.
Someone audits.
What happens when some of those functions are performed by agents?
The answer cannot simply be to treat every agent as another piece of software. Agents can be dynamic. Their behavior can depend on context. They may select tools differently depending on circumstances. They may interact with systems that were never designed to participate in the same workflow. Multiple individually reasonable actions may combine into an outcome nobody explicitly anticipated.
The control environment therefore has to move closer to the action itself.
Enterprises will need to know which agents exist, who owns them, what authority they possess, which systems they can access, what actions they can take, and the boundaries within which they are permitted to operate.
Some actions may be fully autonomous.
Others may require human approval above a financial, operational, legal, or risk threshold.
Some may be prohibited entirely.
And those controls cannot merely exist in policy documents. They increasingly need to be technically enforceable and continuously testable.
This is one reason agentic audit cannot simply become another annual AI risk assessment.
The architecture of governance itself must evolve.
From Periodic Assurance to Continuous Governance
Traditional audit operates largely through sampling.
That approach was born of necessity. Human auditors could not examine every transaction, every control execution, and every decision made across a complex global enterprise.
Machines change the economics of assurance.
If agents generate structured evidence as they operate, organizations can potentially evaluate far more activity continuously.
Instead of asking once a quarter whether a control appears to be functioning, systems can test whether it functioned every time it mattered.
Instead of discovering months later that an agent accumulated excessive permissions, authority can be evaluated before an action occurs.
Instead of waiting for an audit to reconstruct what happened, an enterprise can create verifiable records as the work is performed.
This points toward a fundamental transition:
from retrospective governance to continuous governance.
Governance becomes part of the operating architecture of the enterprise.
Authority can be checked before action.
Controls can be enforced during execution.
Evidence can be generated automatically.
Exceptions can be escalated in real time.
Actions can be recorded in tamper-resistant histories.
Assurance can increasingly operate continuously rather than periodically.
The Stanford program reflects this shift. Friday will begin with The New Audit Mandate, exploring the rise of the agentic enterprise and its implications for board and audit committee oversight. Saturday will focus on Governing Agentic Financial and Control Environments, including ICFR, evidence, assurance, disclosure, accountability, enterprise cases, workshops, peer discussions, and a boardroom simulation. Sunday turns explicitly to Continuous Governance, including agent identity, third-party and supply-chain risk, continuous assurance, and the evolving audit committee operating model.
That progression is intentional.
The challenge is not simply understanding a new technology. It is designing a new governance operating model.
Audit Committees Will Need a New Vocabulary
For audit committees, this will require new questions.
Not simply:
Are we using AI?
But:
Where are agents authorized to act on behalf of the company?
Not simply:
Do we have an AI policy?
But:
How is authority technically enforced?
Not simply:
Have our models been tested?
But:
Can we independently verify the behavior of agents operating inside material workflows?
Not simply:
Who owns AI?
But:
Who is accountable for each agent, each delegated authority, and each consequential action?
And not simply:
Are our controls effective?
But:
Can our control environment operate at the speed and scale of autonomous systems?
These questions cross traditional organizational boundaries.
They connect finance, technology, cybersecurity, internal audit, legal, risk, compliance, operations, and external assurance.
That is precisely why the audit committee is such an important place for this conversation.
Audit committees already sit at the intersection of financial integrity, internal controls, enterprise risk, external assurance, and board-level accountability. In many companies, they will become one of the principal places where the governance implications of agentic systems converge.
The Auditor Changes Too
There is another dimension to the transformation.
Auditors themselves will increasingly use agents.
External auditors will deploy AI to analyze transactions, inspect documentation, identify anomalies, assess controls, generate testing strategies, and examine far larger populations of evidence.
Internal audit teams will do the same.
This may dramatically increase the power of assurance. But it also creates a recursive governance problem.
If an AI agent evaluates another AI agent, what governs the auditor?
How is its methodology validated?
How is independence preserved?
How are hallucinations, model failures, or incomplete evidence identified?
How does an audit committee distinguish between an AI-generated conclusion and independently verified assurance?
The profession therefore faces two transformations simultaneously.
It must learn how to audit agents, while also learning how to audit with agents.
Both require standards, evidence, controls, and accountability.
A New Governance Infrastructure
We believe the companies that navigate this transition successfully will eventually build a new layer of enterprise infrastructure.
Agents will need identities.
Identity will need authority.
Authority will need limits.
Actions will need evidence.
Evidence will need verification.
Exceptions will need escalation.
And consequential actions will need durable records connecting what happened to who or what was authorized to make it happen.
This infrastructure will allow humans and machines to operate together while preserving something every corporation ultimately depends upon: accountability.
Because the central governance question of the agentic enterprise is not whether machines will make decisions.
They already are.
The question is whether organizations will be able to demonstrate that those decisions were authorized, controlled, observable, explainable where necessary, and accountable to human institutions.
The Beginning of Agentic Audit
The first Stanford Directors’ College Agentic Audit Forum will convene a deliberately curated cohort because many of these practices are still being developed.
There is no mature playbook for an audit committee overseeing thousands of autonomous agents.
There is no century of precedent governing machine-scale delegation.
There is no universally accepted standard for agent authority, behavioral assurance, AI-generated audit evidence, or continuous governance.
That makes this a particularly important moment.
The frameworks that govern the agentic enterprise will not emerge from technology companies alone. They will have to be shaped by directors, CFOs, auditors, academics, lawyers, regulators, risk leaders, technologists, investors, and others responsible for the institutions that depend upon trust.
The Forum is intended to help begin that work.
Because the agentic enterprise will require more than better AI.
It will require governance capable of operating at machine speed while preserving human accountability.
And that may ultimately be the new mandate for audit.
The Stanford Directors’ College Agentic Audit Forum takes place October 23–25, 2026 at Stanford Law School and is presented by the Arthur and Toni Rembe Rock Center for Corporate Governance and the Alpha Institute. Participation in the inaugural cohort is by invitation and will be limited. Please sign up here.
Stanford Law School
