Article

Workday AI Governance: How to Deploy Agents Safely

Summarize with
Smiling man in blue shirt standing with crossed arms in office corridor with two colleagues conversing in the background.

Every Workday AI governance conversation eventually comes back to the same question. Before an autonomous agent touches payroll, a candidate pipeline, or a supplier payment, the question must be asked: what happens when it gets something wrong?

And it’s a question most organizations can’t answer it yet. Deloitte’s 2026 State of AI in the Enterprise survey found that only 21% of organizations have a mature governance model for agentic AI, even as most expect to use AI agents at least moderately by 2027.

Mature governance requires clear boundaries for what agents can decide on their own, real-time monitoring of agent behavior, and audit trails that cover every action.

Those three elements are the foundation of a safe deployment. It’s what lets you give agents more room over time, because their autonomy has been earned.

Why Agents Change the Governance Question

Governing AI used to mean reviewing its output. A person asked a question, read the answer, and decided what to do with it. The person stayed accountable because the person took the action.

Agents work differently. Workday’s agents can update records, flag transactions, and move work forward, and more of them are being designed to run in the background without waiting for a prompt. When an agent acts, the question shifts from “Is this answer accurate?” to “Who allowed this action, and how would we know if it went wrong?”

That’s why Workday AI governance is a leadership question as much as a technical one. Your data foundation determines what agents can see and touch, and we cover that groundwork in Your Workday AI Data Strategy Starts Before the First Agent Goes Live. This article picks up where that one ends: governing agents once they’re live. For the bigger picture, see our complete guide to Workday AI.

Governance: Decide Who Decides

Assign Decision Rights Before the First Agent Launches

AI agent governance starts with ownership. For every agent, someone should own three decisions:

  • Approving it for production.
  • Setting how much it can do on its own.
  • Retiring it when it stops earning its place.

Workday’s Agent System of Record gives you a single view of every agent and what it does. It can’t tell you who should hold those decisions. Your organization has to.

In practice, that usually means a small governance group: an executive sponsor, functional owners from HR and Finance, IT and security, internal audit, and legal. The group doesn’t approve every action. It sets the rules agents operate under and revisits them as conditions change.

Match Autonomy to Risk

Agent actions carry different stakes. Answering a benefits question is one thing. Changing someone’s pay is another. A tiered approach keeps autonomy proportional to risk:

  • Recommend. The agent surfaces information or drafts work, and a person decides.
  • Act with approval. The agent prepares the action, and a person approves it before it runs.
  • Act and report. The agent acts within defined limits and logs what it did for review.

Workday lets customers decide which actions an agent takes on its own and which require review. Most organizations should start agents in the lower tiers and move them up as evidence builds. Deloitte found that organizations succeeding with agentic AI follow this measured path: they start with lower-risk use cases, build governance capabilities, and scale deliberately.

Keep People in the Decisions That Affect People

Some decisions should stay with people no matter how well an agent performs. These include hiring, termination, compensation changes, and anything an employee might need to appeal. Workday’s own design points the same way: its Talent Acquisition Agent requires recruiter sign-off before anything goes to a candidate.

AI Security: Vet Agents Before They Start

Agents Face New Kinds of Threats

Traditional application security focuses on who can get into a system. Agents add a new category of AI security risk:

  • Attempts to trick an agent into ignoring its instructions.
  • Attempts to expose those instructions.
  • Leaks of employee data.
  • Unsafe or policy-violating output.

These attacks target how an agent reasons, so network and access controls alone won’t catch them.

Test Before Production, Then Keep Testing

Workday is rolling out Agent Passport to address these risks. Before an agent reaches production, Agent Passport tests it against public standards such as the OWASP LLM Top 10, the NIST AI Risk Management Framework, and MITRE ATLAS. That applies whether Workday built the agent or a third party did. Independent partners sign the results, starting with Cisco, so security teams get an auditable record of what each agent was tested for and who tested it.

The principle holds with or without the tool: no agent should reach production without a documented record of how it was tested. Access still matters for AI security, too. Agents should get only the permissions their job requires, a principle we cover in detail in the data strategy article. Workday’s planned acquisition of C1.ai, which controls what systems employees and agents can reach, shows how central identity has become to keeping agents secure.

Responsible AI: Protect the People Agents Affect

Fairness, Transparency, and Recourse

Workday agents work with some of the most personal data an organization holds: pay, performance, hiring, and benefits. Responsible AI in that setting comes down to three commitments:

  • Fairness. Decisions about people should be fair and checked for bias.
  • Transparency. Employees and candidates should know when AI plays a role in decisions about them.
  • Recourse. People should be able to question an outcome and reach a person who can explain or change it.

The organizations moving most confidently make these commitments formal before launch. BMO, for example, built a responsible AI forum, aligned stakeholders, and brought its policies into Workday before rolling out Workday’s Self-Service Agent to 55,000 employees.

Where Workday Compliance Is Heading

Regulation is moving in the same direction. The EU AI Act classifies AI used in employment decisions as high-risk, with full obligations for those systems now set to apply on December 2, 2027. In the U.S., Colorado replaced its original AI law with a narrower framework, effective January 1, 2027, that centers on transparency and individual rights around automated decisions.

Dates and details will keep shifting, so treat Workday compliance as an ongoing workstream, not a launch checklist. While specific rules settle, the NIST AI Risk Management Framework offers a practical structure for AI risk management. Bring your legal team in early, especially for any agent that touches hiring, pay, or performance.

Monitoring and Controls: See It, Stop It, Prove It

Decision rights and testing set agents up correctly. Monitoring and controls keep them that way. A governed program should be able to do three things at any moment.

See It

You should be able to see what every agent did, what information it used, and who approved it. Workday logs these details for agent actions, and the Agent System of Record brings that activity together so you can confirm agents are working as intended. Assign someone to review it regularly. A log nobody reads protects no one.

Stop It

Controls need to work in real time. Approval gates hold high-stakes actions until a person signs off. Agent Passport adds runtime monitoring that can allow, block, or reroute an agent’s action as it happens. When a problem surfaces, a single revocation can stop or restrict affected agents based on company policy. Decide in advance who has the authority to use that control, and practice using it.

Prove It

Eventually, an auditor, regulator, or employee will ask what an agent did and why. Audit trails, signed test records, and documented approval decisions are your answer. Finance teams already know this discipline well, which is why involving internal audit early makes every later conversation easier.

Monitoring is ongoing work. Workday releases new capabilities regularly, so review your agents after each release. Confirm that permissions, autonomy settings, and test results still match how each agent is actually being used.

A Workday AI Governance Checklist for Executives

Before approving any agent for production, make sure your team can answer six questions:

  1. Who approved this agent, and who owns it now?
  2. What can it do on its own, and what requires human approval?
  3. How was it tested, and by whom?
  4. Who reviews its activity, and how often?
  5. How quickly can we stop it, and who makes that call?
  6. Could we show an auditor exactly what it did and why?

If any answer is unclear, that’s where your AI risk management work begins.

Earning the Right to Let Agents Act

So what happens when an agent gets something wrong? With strong Workday AI governance in place, you have a clear answer. You see the problem quickly, stop it before the impact spreads, and prove exactly what happened and what you changed. That confidence is what allows organizations to give agents more autonomy over time.

The Planet Group helps organizations build that confidence. As Workday’s first global staffing and AMS partner across North America and Europe, we connect teams with specialists from our 400-plus person Workday Certified Network who configure, test, and govern AI agents. Our application management services keep those controls current through every Workday release.

Ready to deploy Workday AI agents with confidence? Talk to our Workday experts.

The Planet Group Logo symbol
Let’s Partner Together
Contact us today for expert talent solutions or career-defining opportunities.