Overwhelmed by AI Governance? Start With These 4 Questions.

Leaders face a growing stack of agentic AI governance frameworks. Underneath the jargon, every one of them asks the same four questions about your AI agents.

When I ask leadership teams which agentic AI governance framework they follow, the answers usually split along geography. Some point to Singapore, some to the European Union (EU) and some to the National Institute of Standards and Technology (NIST) in the U.S. Most assume these frameworks compete, and many conclude governance can wait until the rules settle.

The frameworks do not compete, and governance cannot wait. Underneath the acronyms, every serious framework for AI agents asks the same four questions: what an agent may do, who answers for it, how it can be stopped and how the business proves it. A company that can answer them has done most of what any framework will ask.

The 2026 AI Index from Stanford’s Institute for Human-Centered AI shows why. AI agents completed 66.3% of tasks on a leading computer-use benchmark in 2025, up from about 5% in 2024, while reported AI incidents reached a record 362, up from 233 the year before.

My work advising retail and consumer companies across the Americas, Europe, Australasia and Japan gives me a split-screen view of this debate. The vocabulary changes from market to market, but the questions boards and regulators ask about agents are remarkably similar.

The landscape is smaller than it seems. In January, at the World Economic Forum in Davos, Singapore’s Infocomm Media Development Authority (IMDA) launched what it describes as the first governance framework built specifically for agentic AI. NIST’s voluntary AI Risk Management Framework organizes the work into four functions, and researchers at the University of California, Berkeley published an agentic AI profile built on that structure in February.

Two other pieces complete the picture. ISO/IEC 42001, from the International Organization for Standardization, is the first global standard for an AI management system, and the law sets a floor in some markets. California’s rules on automated decision-making apply to covered businesses starting January 1, 2027, and the EU’s AI Act requires human oversight of high-risk systems, with most of those obligations now applying from December 2027.

1. What an agent may do

Singapore’s framework starts here. It asks organizations to choose suitable use cases first and then limit each agent’s autonomy and its access to tools and data. The World Economic Forum proposed in May that every agent carry an authorization profile recording what it may do, under what conditions and with what oversight.

In practice, that means a written mandate before an agent goes live. It should state what the agent may decide alone, what it may spend or promise and when it must hand a case to a person. A customer service agent might issue goodwill credits up to a fixed amount but hand off any conversation that involves a safety issue or a legal claim.

The mandate should also name what the agent is measured on. Economists Bengt Holmström and Paul Milgrom showed in 1991 that when some duties are easy to measure and others are not, strong rewards on the measurable ones pull effort away from the rest. An agent scored only on speed will optimize speed, and the customer relationship quietly becomes the unmeasured duty.

2. Who answers for it

Every framework insists that accountability stays with people. Singapore calls for checkpoints where human approval is required, and NIST places accountability structures inside its “govern” function.

California’s rules offer the most precise standard. Human involvement there means a reviewer who knows how to interpret and use the technology’s output, analyzes it alongside other relevant information and has the authority to make or change the decision. Covered or not, any business can use that as the job description for an agent’s owner.

One consumer business I advised gave its service agent a generous credit budget and no single owner. Credits crept upward for weeks because each team assumed another was watching. The fix took one meeting: a named owner, a ceiling and a weekly report.

3. How it can be stopped

The EU’s Article 14 states this question most clearly. It requires that the people overseeing a high-risk system can monitor it, override its output and intervene or halt it safely, for example with a “stop” button.

Singapore’s framework calls for technical controls across the agent’s life cycle, including baseline testing and restricting access to approved services. NIST launched an initiative this year that focuses in part on identity and authorization for AI agents.

This matters because an agent using an employee’s login leaves no clear record of who acted. The practical version is simple: Give each agent its own credentials, log every action and test the off switch before launch rather than during an incident. A short weekly review of what each agent decided, escalated and reversed will show whether its limits are too tight, too loose or simply unclear.

4. How the business proves it

The final question is the one customers, auditors and boards eventually ask. ISO/IEC 42001 answers it by treating AI governance as a management system, embedded in everyday processes alongside existing quality and information security systems. Singapore’s fourth dimension adds transparency and training for the people who use agents.

Boards are already asking. More than 62% of directors in the National Association of Corporate Directors’ 2025 survey said they now set aside agenda time for full-board AI discussions. A one-page register listing every agent, its owner, its limits and its last review gives those discussions something concrete to examine.

The first three steps are simple. Build the agent register this quarter, starting with the agents already handling customers or money. Choose one framework as the spine, NIST for its widely used vocabulary or ISO/IEC 42001 when customers expect certification, and borrow Singapore’s agent-specific controls.

Then check the legal floor in each market where the business operates. Agents will keep getting more capable, and the frameworks will keep multiplying. The four questions will not change, and the leaders who can answer them for every agent at work will find that compliance becomes a byproduct of good management rather than a project of its own.

Leaders face a growing stack of agentic AI governance frameworks. Underneath the jargon, every one of them asks the same four questions about your AI agents.

When I ask leadership teams which agentic AI governance framework they follow, the answers usually split along geography. Some point to Singapore, some to the European Union (EU) and some to the National Institute of Standards and Technology (NIST) in the U.S. Most assume these frameworks compete, and many conclude governance can wait until the rules settle.

The frameworks do not compete, and governance cannot wait. Underneath the acronyms, every serious framework for AI agents asks the same four questions: what an agent may do, who answers for it, how it can be stopped and how the business proves it. A company that can answer them has done most of what any framework will ask.