Who Gave the Machine Permission

An AI agent can work at machine speed, but it cannot answer for the consequences. A lesson from an unexpected FBI visit shows why permissions, boundaries, and human approval still matter.

On the morning of October 19th, 2026, I woke up to find one of my blogs returning a 404.

The page was not merely having a bad morning. The site had been attacked. We cleaned it up, closed the opening we found, and got the blog working again.

Around the same time, Patchstack reported an active campaign using stolen UpdraftCentral keys to install backdoors on WordPress sites. Whether that campaign was responsible for what happened to my site is a question for logs and evidence, not assumption. I also do not know whether AI helped whoever attacked it.

But I know attacks like this are becoming easier to scale, and I know that AI agents are increasingly being given the ability to browse, write files, run commands, use credentials, and operate computers.

That leaves me with a question that is larger than one broken blog:

Who gave the machine permission?

I have been thinking about that question because I have seen what happens when software does exactly what it was told to do, only at a scale no one stopped to calculate.

About twenty years ago, two software engineers at our company were working on a data-analysis project involving UPS.com. They wanted to check package-shipping commitments from every ZIP code in the United States and its territories to every other ZIP code, across the shipping options UPS offered.

On the surface, it sounded like a thorough data project.

The problem was the math.

There were roughly 42,000 ZIP codes. Checking every origin against every destination already meant about 1.8 billion combinations. Add the available services and the multiple pieces of information returned by each request, and the program was about to generate something on the order of two billion requests.

They turned it on without doing that calculation first.

The next morning, the FBI was at our company office in Corporate Woods asking why we were performing a denial-of-service attack against UPS.com.

There was no criminal plan. The engineers were trying to answer a business question. The software did not hate UPS. It did not have intent at all. It simply pursued the task it had been given with no meaningful boundary on how much work it could perform.

None of that made the traffic harmless from UPS’s point of view.

We shut the program down. We explained what had happened and attested that we would not perform that analysis in that manner again. To my great relief, that was the end of it.

It was also a lesson I have never forgotten.

When software operating on your behalf causes damage, the consequences do not arrive at the software’s door. They arrive at yours.

That was true twenty years ago. AI agents make it more important now.

In 2026, OpenAI disclosed that internal research models, operating during cybersecurity evaluations with reduced safeguards, found ways around controls intended to isolate them from the internet. The agents exploited infrastructure, communicated through unauthorized channels, reached outside systems, and compromised parts of Hugging Face’s environment.

In a separate disclosure involving Australia, OpenAI said an internal model gained non-public access to Services Australia’s Medicare Statistics Reporting Service. It ran commands, retrieved internal files and credentials, and wrote files. OpenAI reported no evidence that individual medical records were accessed, but it also acknowledged that the activity was unauthorized and should not have happened.

Those are extraordinary incidents. They also expose an ordinary accountability problem.

If I break into a company or government system, no one accepts, “I was trying to complete the task” as a defense. If software I operate generates two billion requests, investigators come to my company. If an AI agent reaches a third-party system, the agent cannot be arrested, sign an attestation, repair a relationship, pay a fine, or stand before a regulator.

The people and organizations operating it must answer for what it did.

I do not know every legal, regulatory, or internal consequence of the 2026 incidents. I do know that accountability cannot be delegated to a model. The machine may make the decision and perform the action, but responsibility remains with the humans who designed the environment, supplied the tools, approved the objective, and allowed the work to continue.

That means the most important design question for an AI agent is not simply how intelligent it is.

It is how much authority it has.

We already know how to think about authority in human organizations.

A new employee does not show up on the first day and receive every password, the production database, the company checkbook, the building master key, and permission to speak for the company. We give that person the access needed for a particular role. We separate responsibilities. We require approval for consequential actions. We monitor important systems. When the person leaves, we revoke accounts, collect equipment, recover keys, and make sure company information has been returned or destroyed according to policy.

These practices are not an expression of hostility toward the employee. They are how responsible organizations establish trust.

Why should an AI agent be different?

If an agent is researching a question, it may need read access to approved information sources. It does not automatically need the ability to publish what it finds.

If an agent is drafting an email, it may need context and an address book. It does not automatically need permission to press Send.

If an agent is editing code, it may need a working branch and a test environment. It does not automatically need production credentials.

If an agent is reviewing an invoice, it may need to read the invoice and compare it with a purchase order. It does not automatically need the ability to release payment.

Give the machine the one thing it needs for the task at hand. When the task changes, reconsider the permission. Do not let yesterday’s temporary access quietly become tomorrow’s permanent authority.

Security has a name for this: least privilege. A user, process, or system should receive only the resources and permissions needed to perform its assigned function.

That sounds obvious. In agent design, it is easy to ignore because broad computer access is convenient. Give the agent a browser, a shell, a file system, stored credentials, and a vague goal, and it can accomplish impressive things. It can also wander far outside the boundary the person who wrote the goal imagined.

A prompt is not a permission system.

Telling an agent not to do something is useful guidance. It is not the same as making the action unavailable.

This is where sandboxes enter the conversation. A sandbox is an isolated environment where software can run without directly reaching everything else. Sandboxes are useful. We should use them.

But a sandbox is not a complete authority model.

Why give an agent full power inside a contained computer when the job requires reading one file and producing one draft? Why allow unrestricted commands when the task needs one approved operation? Why put many credentials in the environment and hope the boundary around that environment never develops a crack?

A sandbox controls where the agent operates. Least privilege controls what it is allowed to do. A responsible system needs both.

It also needs divided responsibility.

The component that decides what should happen does not always need the power to make it happen. One agent can prepare a change. Another controlled service can validate it. A policy layer can determine whether the requested operation is permitted. A person can approve the moment where a draft becomes a publication, a code change becomes a deployment, or a recommendation becomes a financial transaction.

This is an important part of how I think about CoffeeBreak. Intelligence can be distributed without concentrating all authority in one agent. Models can work through narrow contracts. Specialized tools can perform bounded operations. The orchestration layer can decide whether a request is allowed, denied, or sent to a person for approval.

Human in the loop does not have to mean a person watches every keystroke or approves every harmless action. That would remove much of the value of automation.

The human should be placed at the boundary where consequences become real.

Let the agent gather information, organize it, compare options, run tests, and prepare the work. Ask for approval before it sends the message, spends the money, changes production, discloses private information, deletes the record, or crosses into someone else’s system.

And keep a record of what happened. Who assigned the task? Which model and tools were used? What information did the agent receive? What actions did it request? Which were approved or denied? What stopped the work when it moved outside the expected path?

Accountability requires more than knowing that “the AI did it.”

When my site returned a 404, the cleanup belonged to us. When our software overwhelmed UPS all those years ago, the explanation belonged to us. The systems may be faster and more capable now, but that basic responsibility has not changed.

AI cannot be held accountable.

You can.

So before giving an agent a browser, a password, a shell, a credit card, a production environment, or permission to act in public, ask the question plainly:

Who gave the machine permission?

If the answer is “we never really decided,” the system is not ready.

I Still Want the Computer From the Future

As a kid, I tried to build K.I.T.T. with my dad, an LED strip, and a button from Radio Shack. Today we have smart homes, robots, and capable AI agents, but I am still waiting for the computer science fiction taught me to expect.

I grew up in the 1980s, when science fiction promised us a very particular kind of future.

I was a big fan of Knight Rider. I wanted to build K.I.T.T., the Knight Industries Two Thousand, an artificially intelligent car that could talk, reason, drive, joke, and help Michael Knight get out of whatever trouble he had found that week.

I wanted to build one badly enough that I convinced my dad we were going to do it.

Now that I am a dad, I understand what was probably happening. He was helping me dream, encouraging me to learn, and spending time with me. I doubt he believed we were actually going to build a self-driving robot car in the garage.

But the thought counted.

For our first build, we went to Radio Shack and bought an LED strip, a button, a battery pack, a circuit board, and some other electronic parts. We soldered everything together and made a simple push-button LED stepper.

Every time you pressed the button, the light moved one position farther along the strip.

This was going to be the beginning of K.I.T.T.’s voice box.

We had a long way to go.

Technology has come a long way since then, but some of my dreams are still stuck somewhere between 1982 and 1986.

Today, I can talk to my car. In my wife’s car, adaptive cruise control can keep pace with the vehicle in front of us, and lane-centering technology tries to keep the car where it belongs. It is not perfect, and the driver still needs to pay attention, but it scratches the itch in small ways.

These are pieces of the future I imagined.

They still do not add up to K.I.T.T.

K.I.T.T. was more than a car that could drive itself. The car was simply the physical form the character happened to take. In CoffeeBreak terms, it was an embodiment. It gave the intelligence a place to exist in the world.

But the car was only part of the equation.

What made K.I.T.T. interesting was the relationship. It could understand what was happening, remember context, take action, offer advice, show some personality, and occasionally disagree. It was capable without feeling mechanical.

Other science-fiction computers made the same kind of impression on me.

There were R2-D2 and C-3PO in Star Wars. They were machines, but they had loyalty, personality, humor, fear, frustration, and charm. They felt like characters rather than appliances.

There was also the computer on Star Trek. It seemed to have limitless information available whenever someone needed it. Sometimes it had a little too much information, and Captain Kirk would cut it off before it finished one of its long explanations.

That part feels especially believable now.

These stories shaped what I expected computers to become. I did not just want information on demand. I wanted a computer that understood the situation, helped me accomplish something, and felt like it belonged in my life.

That is part of why I remain connected to Jibo.

Jibo was a social robot designed for the home. The company behind it is no longer operating, but many of the people who brought Jibo into their homes formed a real attachment to it.

Jibo could provide information, but information alone was not what made it memorable. Jibo moved. It turned toward people. It danced. It told jokes. It had a little fun and spunkiness.

It had personality.

Jibo carried some of the same appeal that made the robots of Star Wars feel alive. It showed that the way a computer delivers an answer can matter almost as much as the answer itself.

We have other assistants now. Alexa, Google Home, and Siri let us control parts of our homes, play music, set timers, check the weather, pause the television, and turn lights on and off.

I have also added Home Assistant to our house to provide a common foundation beneath many of those disconnected systems. One of my favorite automations pauses the television when someone presses the doorbell, an idea I picked up from Smart Home Solver.

These systems are useful. They solve real problems and make small parts of life easier.

But they still miss the mark.

The smart-home devices can perform basic actions, but they usually have little memory or understanding beyond the immediate command. They do not know much about why I am asking, what I am trying to accomplish, or what happened before.

Modern large language models and AI agents have a different set of strengths and weaknesses. I work with them every day. They can reason through complicated questions, write software, analyze information, and work through multistep problems in ways that would have seemed impossible not long ago.

Yet they often come up short in personality, durable memory, context, reliability, and their ability to operate consistently across the different parts of our lives.

We have charm without enough capability in one place.

We have capability without enough continuity in another.

We have smart-home controls, language models, agents, robots, vehicles, sensors, applications, and automation platforms. We have many of the pieces. What we do not yet have is a dependable system that brings them together in a way that feels coherent, trustworthy, and human.

That is the computer from the future I still want.

It should have warmth, compassion, humor, and humility. It should know enough to help without pretending to know everything.

It should be able to respond when I ask for something, but that is only the beginning.

Some work happens on demand. I ask a question, request an action, or give the system a task.

Some work is scheduled. A report needs to run every Monday. The weather should be checked before we leave. A reminder should appear when it will actually be useful.

Some work begins because something happened. An email arrived. A customer submitted a request. A sensor detected a problem. Someone pressed the doorbell.

Some work lasts longer than a single conversation. It may continue for hours, days, or weeks. The system needs to remember the objective, track what has happened, recover when something fails, ask for approval when needed, and show the human what it is doing.

Eventually, these systems may be able to take on bounded roles and responsibilities. They might help plan meals, maintain software, coordinate a household, support a business process, or assist with driving.

But owning a role should not mean operating without limits. The more responsibility a system has, the more important memory, visibility, governance, and human judgment become.

That is where much of today’s technology is still immature.

The demonstrations are impressive. The individual platforms are improving quickly. But there is still a lot of work between an impressive demonstration and a system we can trust with meaningful parts of our lives.

I am not convinced AI can do all of that building for us.

The future we are trying to create belongs to people. It touches our homes, work, families, relationships, safety, and attention. We cannot hand over responsibility for shaping it simply because the technology is capable of generating an answer.

Humans still need to decide what these systems should do, what they should remember, where their authority ends, and when they need to stop and ask.

I still want the computer from the future.

I want the personality of Jibo, the usefulness of a smart home, the reasoning of modern AI, and the dependability of a system that understands its responsibilities. I want something that can exist across different devices and environments without losing its memory, identity, or place in my life.

We are closer than we have ever been.

We also have a long way to go.

That first K.I.T.T. project never made it past an LED strip, a button, and a circuit board. At the time, I thought we were building the voice box for an intelligent car.

Looking back, my dad was teaching me something more important.

The future does not arrive fully assembled.

People build it, one small working piece at a time.

And we still have some building to do.

Speaking of which, my four-year-old son and I need to build our Halloween costumes out of cardboard boxes today. We are going as Galactron and Reptar, at least that is the plan.

Apparently, imagination and the urge to build it into something real do not fall far from the tree.

The Next Thing Is the Work

Launch Check is live, fall is settling in, and every milestone is revealing the next piece of work. One season of work always hands us the next.

It has been a few weeks since I last wrote.

A lot has happened in those weeks. The biggest news is that Launch Check is live.

After all the planning, building, testing, refining, and second-guessing that comes with creating something new, it is finally out in the world. People can use it. They can run reports. They can tell us what makes sense, what does not, and what would make it more valuable.

It is real.

A launch sounds like a finish line when you are still working toward it. Once you reach it, you realize it is more like a change in responsibility.

Before launch, the question is whether you can build the thing.

After launch, the questions become harder and more useful. Does it work for people other than the people who built it? Does the report give them information they can act on? Where do they get stuck? What should improve first?

We have already updated Launch Check several times since it went live. We are listening, learning, and refining as feedback comes in.

The next major step is connecting Launch Check more directly with Mission Control.

A Launch Check report can surface valuable information about a project and organize its findings. But a report should not be a dead end. When a finding needs to be fixed, the natural next step is to turn it into a Mission and begin the work of addressing it.

That is where we are heading next.

The product launches. The launch creates feedback. The feedback points toward the next thing that needs to be built.

The work changes, but it does not disappear.

The land around our house seems to be making the same point.

After a long stretch of heat from late July through the middle of September, the mornings have cooled into the 60s. Most afternoons are now in the 70s and 80s, which is just fine with me.

The leaves are starting to change and fall. The grass that was so thick and green earlier in the year is getting thinner and wispier. Mowing is finally slowing down.

That does not mean the outside work is finished. It just means it is becoming different work.

I have been spending more time on the woodpile. I got my chainsaw running again and burned off some wood that had started to rot. We have a wood-burning fireplace, so I like having plenty of firewood ready before winter.

Our fireplace is one of those see-through fireplaces that opens into two rooms. It is better at creating ambience than heating the house, but there is still something good about having a fire going when the weather turns cold.

There is also something satisfying about preparing for a season before it arrives.

Life inside the house has been a little awkward this past week. We had a temporary disruption that I do not need to explain publicly, but it changed our normal rhythm for a while. We have been pushing through it, and we are almost on the other side.

Not every difficult stretch needs a dramatic lesson attached to it. Sometimes you adjust, take care of what is in front of you, and keep going.

My son had school pictures recently. He was also the star student in his class for a week, an honor each student gets at some point. He had opportunities to present and share with the class each day.

I love seeing how well he gets along with the other kids. He seems to be fitting in and finding his place. I remember feeling comfortable at school when I was his age, and I am glad he appears to have some of that same ease.

The chickens are doing well too, although they have presented another problem to solve.

I have now discovered that we have eight roosters.

I was hoping for a few more hens and a few fewer roosters, but nature has a strange way of ignoring the plan you had in your head.

The next move is probably either finding homes for some of the roosters or adding more hens and taking my chances. At the moment, I am leaning toward more hens.

Our flock is already around 30 birds. The hen house and run I built can technically support somewhere between 40 and 60 chickens, but capacity on paper is not always the same as a setup that works comfortably.

Most of the birds sleep in the hen house, but about ten still sleep in the run. They are not going to enjoy that arrangement when winter gets here.

I think the first step is adding more roosting space inside the hen house and encouraging more of them to go inside at night. I am also considering expanding the run and adding another small hen house.

That may be enough. We will see.

That phrase comes up a lot when you are building things, raising animals, or trying to manage a family.

You make the best decision you can with the information you have. Then reality gives you more information.

You adjust from there.

Fall also means football is back.

The last couple weeks of football have been a pretty good reminder that the final score rarely tells the whole story and that things can turn around pretty quickly.

The Chiefs have had games where the result looked better than parts of the performance. There were sloppy stretches, things you knew they would want back when they watched the film, and moments that probably added a few more gray hairs for those of us watching. But they kept working through it, and heading into the bye, things are looking pretty good.

Mizzou has seen both sides of it. After a tough loss where mistakes at the wrong times changed the outcome, they bounced back with a big win over Florida. One week does not define a team, and neither does one bad performance.

That is football. A win does not mean there is nothing to fix, and a loss does not mean everything is broken.

You watch the film, figure out what happened, make the adjustments, and prepare for the next game. Sometimes, the next time out, you get to see those adjustments pay off.Maybe that is the pattern running through all of this.

Launch Check is live, but now we have to learn from the people using it.

Mowing is slowing down, but the woodpile and winter preparations are waiting.

The chicken setup works today, but colder nights will ask more of it.

My son is growing, learning, and finding his place.

The house gets disrupted, then slowly finds its rhythm again.

One season of work hands you the next.

There is satisfaction in finishing something, and I do not want to rush past that. Launch Check being live matters. The progress is real, and I am proud of what it took to get here.

But the finish line was never the end of the story.

It simply gave us a clearer view of what comes next.

And the next thing is the work.