AI Could Kill Us All. So What Should Business Leaders Do Monday Morning?

The newest warnings about artificial intelligence deserve serious attention. But businesses and organizations still have decisions to make right now about employees, workflows, data, intellectual property, accountability—and where AI should be allowed into the operation at all.

The latest AI headlines are difficult to ignore.

This week, former Anthropic and OpenAI researcher Jacob Coxon publicly warned that companies developing increasingly advanced artificial intelligence are moving toward self-improving superintelligence without adequate safeguards. His concerns were echoed by current Anthropic researchers, including alignment science lead Evan Hubinger, who stated that he personally believes there is a greater than 10% chance AI could “kill all humans” within the next decade (Luscombe, 2026).

That is an extraordinary statement. It is also important to describe it accurately.

This is not a scientific finding that humanity will disappear within 10 years. It is a personal risk assessment from researchers working closely with advanced AI systems. The larger expert conversation is far from unanimous. While some researchers and policymakers are calling for much stronger controls on artificial superintelligence, others argue that catastrophic scenarios can distract attention from AI risks that are already occurring, including misinformation, employment disruption, cybersecurity vulnerabilities, and other forms of harm (Booth, 2026).

Both things can be true:

The long-term warnings deserve serious attention.

And business leaders still have decisions to make Monday morning.

Your Employees Aren’t Waiting for the AI Debate to Be Settled

While researchers, technology companies, governments, and policymakers debate what advanced AI might eventually become, organizations are already dealing with something much more immediate. And somewhere inside many organizations, sensitive information may already be entering artificial intelligence systems without leadership having fully determined the rules governing that use.

That brings us to a far less sensational—but much more immediate—question:

What are we going to allow AI to do inside our organizations today?

That is where my interest in AI begins.

Not with hype or fear, and not with collecting as many AI tools as possible. I am focused on and teach Operational AI: the controlled use of artificial intelligence to improve how work moves through an organization without surrendering human judgment, accountability, intellectual property, data stewardship, or trust.

That distinction matters because responsible AI adoption is not simply a technology issue. The National Institute of Standards and Technology’s Artificial Intelligence Risk Management Framework emphasizes that organizations need toconsider trustworthiness, risk, context, governance, and ongoing management throughout the AI lifecycle—not merely at the moment a tool is purchased (Autio et al., 2024).

We Are Having Two AI Conversations at Once

The first is the frontier AI conversation. That conversation includes superintelligence, alignment, loss of human control, cybersecurity, regulation, autonomous systems, catastrophic risk, and the long-term consequences of increasingly capable artificial intelligence.

Those are serious questions. They require researchers, technologists, policymakers, ethicists, legal experts, and others with specialized expertise.

The second is the operational AI conversation. That conversation sounds considerably less futuristic:

• Should an employee put this document into an AI system?
• Who checks its work?
• What happens when it is wrong?
• Which decisions must remain human?
• How will we know whether AI actually saved us time or money?

And perhaps most importantly:

Who is accountable for what happens next?

Most business leaders cannot control the global race toward increasingly powerful artificial intelligence. They absolutely can control how AI enters their organizations.

You do not have to solve the future of artificial intelligence before deciding whether someone should paste a customer file, unpublished intellectual property, employee record, financial report, confidential strategy document, or proprietary process into an AI platform.

That is a leadership decision. And increasingly, it is a governance decision. The National Institute of Standards and Technology’s (NIST) guidance specifically encourages organizations to manage generative AI according to their goals, requirements, resources, and risk tolerance rather than treating every use case identically (Autio et al., 2024).

AI Should Have to EARN Its Place in Your Operation

One of the biggest mistakes I see in conversations about artificial intelligence is starting with the tool.

“What can ChatGPT do?”

“What can Claude do?”

“What AI platform should we buy?”

“What should we automate?”

I would start somewhere else.

What problem are we trying to solve?

Before AI gets access to a workflow, I believe that use case should have to EARN its place in the operation.

E — Existing Problem

• What specific operational problem are we solving?

Start with the friction. Otherwise, you can spend money implementing artificial intelligence into a process that never needed AI in the first place.

Technology added to a broken process often gives you a faster broken process.

A — Authorized Inputs

Can this information appropriately enter the AI system being used?

That question gets skipped far too often. NIST’s generative AI guidance identifies data privacy, information security, intellectual property, and other risks as considerations organizations may need to manage when implementing generative AI systems (Autio et al., 2024). If the AI requires information it should not have, it has already failed the test.

R — Reviewable and Reversible Output

Can a qualified human reasonably verify what AI produces? And can an error be caught before it creates material consequences?

This is critical because an output that looks polished can still be wrong.

AI can organize information beautifully. It can also organize incorrect information beautifully. Those are not the same thing. The higher the consequence of an error, the stronger the human review and organizational controls should become.

N — Named Human Owner

Who owns the outcome?

Not the software. Not “the AI.” A person.

• If AI drafts a communication, who approves it?
• If it identifies a problem in a report, who verifies it?
• If an automated workflow takes an action, who is responsible for monitoring it?
• If something goes wrong, who has authority to stop the process?
• If nobody owns the output, AI does not own it either. You have an accountability problem.

That is why responsible AI adoption is not primarily a prompting problem. It is an operational problem.

Consider a Very Ordinary Meeting

Imagine your leadership team holds a weekly meeting.

The meeting lasts 60 minutes.

Someone then spends another 45 minutes reviewing notes, identifying decisions, extracting assignments, organizing deadlines, and creating the follow-up communication.

Could AI reduce some of that administrative burden? Possibly.

An approved AI system might help produce a first structured draft:

Decision → Action → Owner → Deadline → Follow-Up Required

Count the review time. Count the corrections. Count the rework.

If you do not, you may call something “AI efficiency” that simply moved the labor somewhere less visible.

Stop Making Your Best Employees Human Filing Cabinets

Your best people should not spend part of every day functioning as human filing cabinets, transcription services, search engines, or copy-and-paste machines.

And Yes, Employees Are Right to Ask About Their Jobs

Organizations should stop pretending this question does not exist.

Some tasks and roles will change. And some organizations will undoubtedly use artificial intelligence primarily as a labor-reduction strategy. Pretending otherwise does not build employee trust.

But there is another question responsible leaders can ask first:

Where are skilled people spending time doing work that does not require their skill?

There is an enormous difference between:

“How many people can we eliminate?”

and

“What unnecessary friction can we remove from the work our people are already doing?”

One is primarily a headcount strategy. The other is an operational-capacity strategy. I am far more interested in the second. The goal should not automatically be removing the person, but removing unnecessary friction from the person’s work.

AI Should Support Decisions—Not Quietly Become the Decision-Maker

There is another distinction leaders need to make. AI can be useful for decision support, but that does not mean AI should own the decision.

My rule is simple:

Do not ask AI to decide what matters. Give it criteria your organization has already decided matter.

Let technology help organize information and let humans exercise judgment.

Not Every AI Use Case Deserves the Same Level of Concern

Organizations need more sophisticated policies than either:

“Nobody can use AI.”

or:

“AI is great. Everybody figure it out.”

Think instead in levels of risk. A low-risk use involving approved, nonconfidential information that a human can easily verify may require relatively simple controls.

A use involving financial data, employee information, contracts, customer records, unpublished intellectual property, or proprietary strategy deserves significantly more scrutiny. And some uses may simply be inappropriate.

The question is not:

“Do we use AI?” That question has become almost meaningless.

The better question is:

“Which AI use cases do we permit, under what conditions, using which systems, with what data, subject to what human review?”

That is governance.

• Policy before prompt.
• Workflow before tool.
• Evidence before scale.

Publishing Makes These Questions Especially Visible

These are not publishing-only issues. They affect professional-services firms, nonprofits, associations, educational organizations, agencies, corporations and small businesses.

But publishing provides an especially revealing example because our industry sits at the intersection of technology, intellectual property, human creativity, confidential manuscripts, contracts, metadata, production systems, author relationships and commerce. And copyright deserves particular attention.

In its report on the copyrightability of generative AI outputs, the U.S. Copyright Office concluded that material generated with AI may receive copyright protection when sufficient human-authored expressive elements are present, but that merely providing prompts does not by itself supply the necessary human authorship. The Office also emphasized that using AI as an assistive tool does not automatically prevent copyright protection for a larger human-authored work (U.S. Copyright Office, 2025).

For publishers, I make an additional distinction when I talk about Operational AI:

I am not advocating for AI to replace authorship, editorial judgment, or the human intellectual content publishers exist to bring into the marketplace. My interest is in what happens around that content.

• The repetitive administrative work.
• The places where talented publishing professionals are losing capacity because the underlying operation requires too much unnecessary manual effort.

That is where the conversation gets interesting.

This Is the Conversation I’m Bringing to IBPA on September 23rd

On September 23rd, I will teach Operational AI: Four Ways to Build a Smarter, More Efficient Publishing Operation during the Independent Book Publishers Association’s virtual Boot Camp for Staff of Independent Publishers.

Rather than giving publishing professionals another broad overview of artificial intelligence, I will focus on four operational applications:

1. Reducing repetitive work.
2. Strengthening internal systems.
3. Improving communication and decision support.
4. Extending the capacity of the people already doing the work.

And all four will be examined through human oversight, responsible implementation, ethical standards, and respect for intellectual property. IBPA’s September 23 virtual Boot Camp is designed for early-career publishing professionals who want to contribute more strategically to their organizations (Independent Book Publishers Association [IBPA], 2026).

That audience makes this conversation particularly important. AI implementation should not happen only in the executive suite. The people closest to the actual work often know exactly where the friction lives. They also know where a poorly designed solution will create three new problems while supposedly fixing one.

If you work in publishing, manage publishing professionals, or lead an independent publishing organization, I hope you will join us on September 23rd (click here).

And If You Lead an Organization Outside Publishing…

Do not dismiss this as a publishing conversation.

• Change the documents.
• Change the workflow.
• Change the customer.
• Change the intellectual property.

The management questions remain remarkably similar. Those are Operational AI questions. And every organization adopting artificial intelligence will eventually have to answer them.

The smartest organizations will answer them before they become AI-enabled workplaces by accident.

The Bigger AI Question Does Not Excuse Us From the Immediate One

The warnings about advanced AI deserve serious attention. But leaders still have decisions to make today.

• Govern what you can govern.
• Protect what needs protecting.
• Measure what you implement.
• Train your people.
• Keep humans accountable.
• And make AI prove that it deserves a place inside your operation.

Because the objective is not to become an AI-powered organization. The objective is to build a stronger organization that knows exactly when AI deserves a role—and when the answer should simply be no.

Alesha Brown, CEO, Fruition Publishing Concierge Services®

Editor-in-Chief, Published! Magazine®

Publisher| Business Strategist | Operational AI Consultant

About Alesha Brown

Alesha Brown is an award-winning entrepreneur, publisher, speaker, and business strategist whose work focuses on operational efficiency, sustainable business growth, and responsible AI implementation. She is the Founder and CEO of Fruition Publishing Concierge Services® and a former four-year member of the Independent Book Publishers Association Board of Directors.

Her Operational AI work focuses on helping leaders identify where artificial intelligence can strengthen workflows, systems, communication, and organizational capacity while preserving human judgment, accountability, and responsible governance.

Alesha speaks and trains for businesses, associations, leadership teams, and publishing organizations on Operational AI, business systems, and strategic implementation.