AI agents are becoming part of everyday business operations. But as they gain access to company information and systems, leaders need to answer one important question: Who is accountable for what they do?

AI is moving beyond answering questions and generating content. Agents can now interact with business systems, access information, and carry tasks forward.

Recent announcements from Microsoft, OpenAI, and Google show how quickly this is changing. At the same time, real-world incidents remind us why access, confidentiality, and human review matter.

For business leaders, the message is simple:

AI needs more than access. It needs clear ownership, boundaries, and human accountability.

1. Every AI agent needs an owner

Microsoft’s November 17 announcement introduced expanded capabilities for managing and governing AI agents through Agent 365.

For businesses, this raises an important question: Do you know which agents are operating in your organization and who is responsible for them?

Before deploying an agent, establish:

  • What is it supposed to do?
  • What information can it access?
  • Who is responsible for monitoring it?
  • When should its access be reviewed or removed?

An AI agent should never operate without a clearly defined purpose and accountable human owner.

Source: Microsoft Agent 365 — Ignite 2026

2. More access means more responsibility

OpenAI made personal dots available by default in eligible Enterprise, Edu, and Healthcare workspaces beginning November 18, 2026.

These agents can continue working on tasks between conversations, depending on their configuration and permissions.

That creates opportunities for businesses, but also makes permission management more important.

Leaders should understand:

  • Who can create and use agents?
  • Which applications and information can they access?
  • Who reviews their activity?
  • How can access be stopped when it is no longer needed?

Allowing employees to use AI does not mean giving every agent unrestricted access.

Sources: OpenAI Workspace Release Notes · Managing Dots in ChatGPT Workspaces

3. Protect information. Verify the output.

Two November investigations published by the U.S. Department of Justice highlighted serious concerns involving AI use.

One involved confidential information being uploaded to a public AI platform. Another involved fabricated legal quotations and inaccurate case references in court filings.

The lesson applies beyond legal work.

Employees need to know what information they can share with AI and how to check what AI produces.

Every business should establish:

  • Which AI tools are approved.
  • What company, client, and employee information can be entered.
  • Who checks AI-generated work.
  • When an output must be rejected or escalated.

AI can support professional work, but people remain responsible for what they approve and use.

Sources: DOJ Investigative Summary 5 · DOJ Investigative Summary 6

4. What happens when your technology stops working?

The Microsoft 365 service disruption in November was a reminder of how dependent businesses have become on connected technology.

Email, documents, meetings, and other essential services can rely on the same provider.

If a major platform becomes unavailable, can your team continue serving customers and handling essential work?

Consider:

  • How will employees communicate?
  • Where can they access essential information?
  • Which processes can continue manually?
  • Who decides when the backup plan is activated?

AI integration should strengthen business operations, not make them more fragile.

Sources: Microsoft 365 Incident Information · Associated Press — Microsoft Services Outage

5. Keep up with AI provider policies

AI providers regularly update their usage policies.

Anthropic’s updated policy took effect November 12, 2026, and OpenAI’s updated policies took effect November 20, 2026.

Businesses should understand the rules attached to the tools they use.

That includes knowing which activities are restricted, what safeguards are required, and where human review must remain part of the process.

Don’t assume that because a tool can perform a task, it is automatically appropriate for your business to use it that way.

Sources: Anthropic Usage Policy · OpenAI Usage Policies

6. Automation still needs human approval

Google Cloud introduced new capabilities for managing organization policies through AI-assisted tools.

These developments can help teams inspect settings and work with policy configurations more efficiently.

But faster administration does not mean every change should happen automatically.

A useful approach is to let AI help identify issues, explain options, and prepare recommendations while people retain approval authority over consequential changes.

Technology can help make decisions easier. It should not remove the accountability behind them.

Source: Google Cloud — Organization Policy Updates

What should business leaders do next?

You don’t need to adopt every new AI capability as soon as it becomes available.

Start by asking six practical questions:

  1. Do we know which AI tools and agents are being used?
  2. Is every agent connected to a clear business purpose and human owner?
  3. Are our information and access boundaries defined?
  4. Do employees know when and how to review AI-generated work?
  5. Can our essential operations continue if a technology provider becomes unavailable?
  6. Do we regularly review whether our AI practices are still appropriate?

If you cannot answer these questions confidently, it may be time to strengthen your foundations before expanding AI use.

Lead AI with people at the center

At Get Desky, I approach AI integration through four connected responsibilities:

Govern. Establish standards, permissions, and accountability.

Prioritize. Identify the business problems worth solving.

Operationalize. Build practical workflows with clear ownership and human review.

Normalize. Help people use AI consistently and responsibly.

AI should make a business stronger without taking people out of the decisions that matter.

The goal isn’t simply to introduce more technology. It’s to build better ways of working, with people who understand their responsibilities and have the capacity to do meaningful work.

An AI agent can hold access. Only a human can hold responsibility.


Jennifer G. Bennett is the founder of Get Desky and a Fractional Chief AI Officer. She helps business leaders connect AI decisions to business priorities, practical systems, and responsible team adoption.

Explore Get Desky

Sources & Further Reading

Product features, availability, and policies may change. Refer to the linked provider documentation for current details.