When the Hotel’s AI Acts, Who Is Accountable? Building a Hospitality Agent Governance Architecture

Regulators have begun to treat AI that acts differently from AI that assists. Singapore published what it described as the first-of-its-kind governance framework specifically for responsible agentic AI deployment in January 2026 and updated it in May. China issued its first national policy on AI agents in May, and in September its national cybersecurity standards committee (TC260) made agents a distinct risk category in its AI Safety Governance Framework 3.0.

For hospitality, this raises a governance problem that most corporate AI policies are not built to solve. I use the term delegated digital authority to describe the authority an organization intentionally grants an autonomous system to access information, make decisions, initiate transactions, communicate externally, or take actions on its behalf within defined boundaries. Once an agent holds that authority, the central question is no longer whether it can perform a task. It is: who authorized it to act, how far does that authority extend, and who remains accountable for the outcome?

One Hotel, Multiple Sources of Authority

A hotel displaying one brand may be owned by one company, operated by another, franchised under a third, and connected to technologies supplied by many more. Consider a guest-facing agent handling a reservation. Its workflow may touch the property management system, central reservations, OTAs, revenue management, payments, loyalty, guest profiles, housekeeping inventory and outside partners.

Authority over those systems is rarely held in one place: the property controls certain guest-service decisions, the management company sets operating procedures, the brand owns reservation infrastructure and loyalty rules, ownership sets financial limits and risk tolerance, and vendors control the systems themselves. The first mistake to avoid is assuming that because an agent can technically reach several systems, one party can authorize everything it does. Technical access and organizational authority are not the same thing.

Governance Should Follow Existing Decision Rights

Agent governance should begin by mapping the hotel’s existing decision rights rather than inventing new ones:

Seen this way, the governance question is not simply whether an agent is authorized. It is: authorized by whom, to do what, within which system, and on whose behalf?

Cross-System Workflows Create the Real Risk

Take service recovery. A guest complains about a room. An agent might access the guest profile, review the reservation, check loyalty status, check inventory, offer compensation, modify the folio, reassign the room, and create a housekeeping or engineering task. To the guest, this looks like one response. In governance terms, it crosses multiple systems, permissions, departments and, potentially, organizations.

“Resolve guest complaints” is therefore too broad to be a permission. May the agent view loyalty information? Move a guest? Issue compensation? Alter the folio? Open a maintenance work order? Offer a future-stay credit? Contact the guest externally? Each action may have a different owner and a different threshold.

Physical Operations Need Different Boundaries

An engineering agent that detects unusual HVAC performance and opens a work order carries relatively limited risk. Allowing the same agent to shut down equipment, alter building-management controls, change access-control settings, or affect fire and life-safety systems is a different matter entirely.

The same applies to housekeeping. An agent may reasonably prioritize rooms by arrivals, departures and operational need. But once its optimization assigns overtime, sends employees home, changes staffing levels, or systematically routes less desirable assignments to certain people, it has crossed from task optimization into workforce management. The capability is technical. The consequence is managerial.

Human Resources Requires Consequence-Based Governance

In HR, automating interview scheduling or onboarding reminders is low-stakes. Candidate ranking, employment screening, schedule changes that affect income, performance evaluation, discipline, promotion and termination carry progressively greater human consequences. Here the relevant question is not whether the AI can make the decision, but whether the organization should delegate it.

Regulation is moving in the same direction. The EU AI Act lists as high-risk the AI systems used for recruitment and selection, and for decisions on promotion, termination, task allocation based on individual behavior or traits, and performance monitoring. Under the Digital Omnibus amendments approved this year, those obligations now apply from 2 December 2027. Singapore’s framework likewise calls on organizations to bound agent autonomy, keep humans meaningfully accountable, and “define significant checkpoints or action boundaries that require human approval.” Those principles translate directly into hospitality HR decision rights.

External Agents Change the Governance Equation

Hospitality companies must also prepare for agents they do not own. China’s policy already anticipates this, with provisions on agent digital identity, capability declarations, trusted interconnection and compliant payments. Notably, it also names lodging among the commercial venues where agents should improve operating efficiency (Sections 4 and 24).

Imagine a corporate traveler’s AI negotiating directly with a hotel’s agent. The hotel would need to know:

• whether the agent is genuinely authorized by the traveler;

• what it may do, and within what spending limit: access loyalty data, cancel a non-refundable booking, authorize payment;

• what happens when its instructions conflict with corporate travel policy; and

• who resolves the dispute if the two agents reach the wrong outcome.

These are not simply cybersecurity questions. They are questions of commercial authority and accountability.

A Hospitality Agent Governance Architecture

Drawing on the Singaporean and Chinese frameworks, and on how hotels are actually owned and operated, I propose seven components.

1. Identity. Every agent should have a defined identity, purpose, business owner and operational sponsor. Both China’s policy and TC260’s Framework 3.0, which lists identity and permission misuse among its core agent risks, treat identity as foundational.

2. Authority. An authority matrix should define the systems, data, transactions, financial thresholds, decisions and external communications each agent may use or initiate, and what it is prohibited from doing.

3. Human control. Organizations should distinguish among autonomous actions, human-authorized actions and human-reserved decisions. This three-tier structure closely mirrors China’s own distinction between decisions reserved to the user, decisions requiring the user’s authorization, and decisions an agent may make autonomously (Section 6). Organizations should also decide who can override, suspend or disable an agent.

4. Data access. Least privilege applies. An agent should not receive broad access to guest, payment, employee, loyalty or security data simply because more data might improve its performance.

5. Traceability. Significant actions should be reconstructable: what the agent did, what information it used, which systems or tools it accessed, whether another agent was involved, and whether a human approved.

6. External-agent governance. Rules for agents outside the hotel’s control, covering identity, authorization, data exchange, transaction limits and dispute resolution.

7. Accountability. Every autonomous workflow should lead back to an accountable organizational role. An organization may delegate execution; it should not assume that delegation eliminates responsibility.

Where Does Accountability Sit?

This may be the hardest hospitality-specific issue. Suppose a guest-facing agent uses a brand reservation platform, a third-party AI model, a property management system, a payment provider, and operating parameters set by the management company, and it makes a decision that harms the guest. Does accountability sit with the owner, the management company, the brand, the technology vendor, or the manager responsible for oversight?

Early case law offers a useful warning that businesses may not be able to distance themselves from customer-facing AI simply by pointing to the software. In Moffatt v. Air Canada (2024), a British Columbia tribunal held the airline liable for incorrect fare information given by its website chatbot, rejecting Air Canada’s argument that it could not be held liable for what the chatbot said. It was a small-claims decision, but it illustrates a practical reality: from the customer’s perspective, the agent often is the company.

My answer has three parts. Externally, the first accountability question will often focus on the organization presented to the guest as responsible for the interaction, although contractual and legal liability can vary substantially by jurisdiction and ownership structure. Internally, it should be allocated before deployment rather than litigated after an incident: management agreements, franchise agreements and technology contracts should state who may grant an agent authority, who sets its limits, who monitors it, and who bears the cost when it errs. Operationally, every agent should have one named accountable sponsor, normally the role that configures what the agent is permitted to do.

Agentic AI Is a Corporate Governance Question

This is why agentic AI cannot sit solely within IT. Depending on the use case, governance may involve operations, HR, finance, revenue management, legal, risk, ownership, brand leadership and technology partners. Boards already govern delegated authority among people through spending limits, signing authority, access privileges and reporting lines. Agentic systems add a new kind of actor to that structure.

Hospitality organizations preparing for agentic AI should be able to answer seven questions. Which agents are operating? Who owns each one? Whose authority are they exercising? What decisions can they make? What systems and data can they access? Who can stop them? And who is accountable when something goes wrong?

Good agentic AI governance does not require hospitality to reinvent how it governs authority. It requires extending that discipline to digital actors. The next governance challenge for hospitality is not teaching AI how to act. It is deciding how much authority we are willing to give it.