Finding Truth: When Machines Learn to Kill, What Do We Do Next?

Every now and then, a story comes along that makes anyone who calls themselves a student of truth pause for a minute. Not just to react to the headline, but to sit with it. To ask what it means today, what it could become tomorrow, and what kind of responsibility comes with knowing.

This week, that story is about autonomous weapons, or “killer robots,” as the headline writers like to call them. According to New Scientist, drone-maker Alexander Kokhanovskyy said a one-time Ukrainian battlefield test used fully autonomous drones without human oversight, and later assessment credited the drones with killing Russian soldiers and destroying a truck.

If true, that matters. A lot. It means a line many people hoped we would never cross may already be behind us. But it also means we are being given a chance to respond with wisdom while the rules are still being written.

And if we’re going to be honest seekers of truth, we have to sit with that for a minute.

What the article actually says (and what it doesn’t)

The article lays out a few key facts:

·         A senior figure in the Ukrainian defense industry says autonomous weapons were used in a battlefield test that killed soldiers without human oversight.

·         The technology has existed for years, and history tells us humans don’t invent weapons and then politely refrain from using them.

·         The UN has been trying, unsuccessfully, to ban fully autonomous weapons for over a decade, with several major nations blocking progress.

·         And perhaps the most unsettling detail is this: these systems can be built from inexpensive parts ordered online and open-source software.

·         The article even says a “tech-literate teenager” could build one.

That last point is where my coffee got cold.

Because once a technology becomes cheap, accessible, and modular, it stops being a geopolitical problem and starts being a societal one.

What makes this moment different is not just that the technology exists. It is that the ingredients are getting cheaper, smaller, smarter, and easier to connect. That is when a military question starts to become a public-safety question.

One reason this gets confusing is that there is not one neat, globally agreed definition of a lethal autonomous weapon. The United Nations Office for Disarmament Affairs says there is no common definition yet, but describes the concern broadly as weapons that can use autonomous functions to detect, select, or engage targets after activation. It also notes that not every autonomous weapon needs advanced AI. Some systems follow predefined rules, while others may use AI to interpret sensor data and adjust to changing conditions. [Source: United Nations Office for Disarmament Affairs]

That distinction matters. A system can be automated without being truly intelligent. It can be dangerous because of speed, persistence, and delegation, even before we get to science-fiction ideas of machines “thinking” for themselves.

The battlefield implications are obvious, but the civilian ones may be worse

Let’s start with the military side.

Militaries are interested in these systems for reasons that are not hard to understand. They can scout ahead of troops, loiter over a battlefield, search for radar or armored vehicles, support reconnaissance, and sometimes strike quickly when a target appears. The United Nations describes loitering munitions as weapons that can wait in a predefined area until a target is found by an operator or onboard sensors, then attack. That is very different from a traditional missile that simply flies to a fixed point. [Source: United Nations Office for Disarmament Affairs]

The International Committee of the Red Cross has warned that these systems raise a special problem because, after activation, the person using the weapon may not know the exact target, timing, or location of the eventual strike. That is where the moral weight shifts. The human starts the process, but the machine may finish it in circumstances the human did not fully see. [Source: International Committee of the Red Cross]

Autonomous weapons change the nature of warfare in three ways:

1.      Speed. Machines don’t hesitate. They don’t get tired. They don’t wait for permission.

2.      Scale. If one drone can kill autonomously, a thousand can do it faster than a human commander can blink.

3.      Accountability. When a machine makes a lethal mistake, who is responsible? The programmer? The commander? The algorithm? The nation?

These are the questions ethicists have been warning about for years, and they’re not wrong. The article points out that autonomous weapons “deprive combatants of their dignity” and make war too easy to wage.

There is also a proliferation problem. Researchers and policy analysts have noted that loitering munitions and autonomous-capable systems are spreading quickly. That does not mean every system is fully autonomous, but it does show how quickly battlefield habits can become marketplace products, doctrine, and expectation. [Sources: United Nations Office for Disarmament Affairs; American Society of International Law]

But here’s where I want to take this conversation a little further.

Because the battlefield isn’t the only place this technology can go.

What happens when this tech leaves the battlefield?

The public-safety concern is not limited to war. It reaches into airports, schools, stadiums, political events, houses of worship, and anywhere people gather. In a country already struggling with mass violence, that should make us pause.

Civilian drones already do a lot of good work. They inspect power lines, help farmers, support search and rescue, map storm damage, and give first responders eyes in places that may be unsafe for people. That is part of why this conversation is hard. The same basic platform can be useful, harmless, careless, or dangerous depending on the person using it and the guardrails around it.

Public-safety agencies are already being asked to think this way. CISA’s “Be Air Aware” materials warn that unmanned aircraft can create cyber and physical risks for critical infrastructure and public gatherings, and the agency has issued guidance on suspicious drone activity, detection technology, and safe handling of downed drones. That is not future fantasy. It is already part of the risk planning for critical infrastructure and large events. [Source: Cybersecurity and Infrastructure Security Agency]

I want to be careful here. This is not a how-to conversation, and it should never become one. But we can still speak plainly about the broader truth. Low-cost drones, cameras, software, and automated decision-making are becoming easier to combine in dangerous ways.

That changes the public-safety conversation. An automated system can put distance between the person who launches it and the harm it causes. It can keep moving after the person is gone. It can act faster than people can react. And if we wait until the first tragedy to begin thinking seriously, we will already be behind.

The danger is not only the dramatic worst-case scenario. It is also the messy middle: a drone near an airport runway, a stadium delay that triggers panic, surveillance over a power facility, a package dropped into a prison yard, or a confused local responder who does not yet have the authority or tools to act safely.

That does not mean we should panic. It does mean we should be honest. A tool that can fly, sense, decide, and keep operating after the person behind it has stepped away changes the way we think about prevention, response, and responsibility.

That is the uncomfortable part of this story: once a weapon becomes automated, portable, and cheap enough, the danger is no longer limited to battlefields or governments.

And that’s the part of this story we have to talk about, not to scare ourselves, but to be honest about the world we’re walking into.

How hard would it be to build a home‑grown autonomous weapon?

Let’s be clear: I’m not going to describe how to build anything harmful. But we can talk about feasibility in broad terms.

The article says these systems can be built from inexpensive parts and open-source software. That may be true, but “inexpensive” doesn’t mean “easy.”

At a high level, a dangerous autonomous system would require several things working together: mobility, sensing, decision-making software, power, stability, and some way to cause harm. Each of those pieces has legitimate civilian uses. The problem is what happens when they are combined with bad intent.

That’s not trivial. It’s not something the average person could do. It’s not something a teenager could do well without serious risk to themselves.

But is it possible? Yes. And that’s the part we can’t ignore.

Because the barrier to entry is dropping every year.

That is why the civilian-misuse conversation cannot wait for a perfect villain or a perfect case study. We already know enough to see the direction of travel: smaller systems, better sensors, more automation, cheaper components, and more people with access to the pieces.

If we can’t put the genie back in the box, what can we do?

This is where the coffee conversation gets practical. We may not be able to uninvent autonomous systems. Once a powerful technology exists, someone will keep improving it, selling it, hiding it, or using it. So maybe the better question is not, “Can we put the genie back in the box?” Maybe it is, “Can we build a safer box around it?”

I think we can. Not perfectly. Not magically. But seriously. Here are a few places to start:

What would a safer box look like?

The answer probably is not one big rule. It is layers. Better design standards. Better laws. Better detection. Better public awareness. Better accountability. No single layer catches everything, but every layer makes careless or malicious use harder.

My starting point is simple: the more power a machine has over human life, the more human responsibility must surround it.

The United Nations Secretary-General has said human control over the use of force is essential, and the International Committee of the Red Cross has urged legally binding rules that would prohibit systems whose effects cannot be understood, predicted, or explained. That gives us a useful starting point: if humans cannot meaningfully understand and limit what the system will do, society should not treat deployment as acceptable. [Sources: UN News; International Committee of the Red Cross]

·         Keep humans legally responsible. If a machine is used to take a life, a human chain of command should still be accountable. “The algorithm did it” cannot become a moral escape hatch.

·         Mandate meaningful human control. For lethal use, society can require a human decision-maker at the final point of force, with clear standards for what “meaningful” actually means.

·         Test before deployment. Any system that can act with lethal consequences should go through independent safety testing, red-team review, bias testing, failure-mode analysis, and regular recertification. If a company or agency cannot explain how the system behaves when sensors fail, maps are wrong, signals are jammed, or civilians are nearby, then it should not be fielded.

·         Require audit trails. Systems that identify or engage targets should keep records of decision logic, sensor inputs, approvals, and overrides so investigators can reconstruct what happened afterward.

·         Build in technical locks. Think of it like requiring brakes, headlights, seatbelts, license plates, and insurance before we let cars share the road. For autonomous systems, the equivalent might be geofencing that keeps drones away from schools, stadiums, airports, government buildings, and critical infrastructure; automatic return-to-home when a system loses authorization; and remote disable functions when a system enters a restricted area.

·         Require identity and traceability. We can require remote identification, signed firmware updates, tamper detection, and logs that cannot simply be erased after something goes wrong. If a system causes harm, investigators should be able to trace who owned it, who updated it, what it was allowed to do, and what decisions were made.

·         Set hard safety limits. That could include payload limits, hard-coded no-fly zones, restricted operating modes, automatic shutdown when an authorized operator cannot be verified, and design requirements that make dangerous modification harder. None of those controls is perfect on its own, but together they raise the cost of misuse.

·         Regulate dangerous combinations, not just parts. Drones, cameras, software, batteries, and mounts may each be legal on their own. The law has to pay attention to how pieces are combined into a weaponized system.

·         Harden public spaces. Airports, stadiums, schools, and critical infrastructure may need better detection, alerting, and safe counter-drone response plans.

·         Prepare local responders. Local police, emergency managers, airport operators, school districts, and event security teams are often the first to see a drone problem, but they may not be the ones legally authorized or technically equipped to stop it. A safer society needs training, clear handoff procedures, and lawful tools that match the speed of the threat.

·         Use purchasing power. Governments can refuse to buy systems that do not meet safety, testing, logging, and human-control standards. That can move industry faster than speeches.

·         Create international red lines. Even if a total ban fails, nations can still agree on narrower rules: no autonomous nuclear launch, no autonomous attacks on civilians, no systems that cannot be recalled, and no deployment without review.

Internationally, there is movement, even if it is slow. Recent United Nations discussions have focused on a two-part approach: banning the most unacceptable systems and regulating others with strict limits. That feels like common sense, and it is also a hopeful sign that nations are at least beginning to name the danger together. Some uses may be too dangerous to permit at all. Others may require narrow rules about where, when, how long, under whose supervision, and against what kinds of targets they can operate. [Sources: UN News; American Society of International Law]

But controls alone are not enough. Technology can help, but culture has to carry some of the weight too. We need public education that teaches people what drones can and cannot legally do. We need clear reporting channels when someone sees suspicious drone activity near a school, stadium, airport, prison, power facility, or public event. We need emergency plans that local police, schools, hospitals, churches, utilities, airport operators, and event organizers actually practice. We need manufacturers, retailers, software platforms, and government agencies sharing responsibility instead of pointing fingers after something goes wrong.

And yes, we also need civic courage. That means lawmakers who understand the technology well enough to regulate it, citizens who can talk about safety without turning every conversation into panic, and companies willing to build guardrails before the lawsuit, not after it.

What can ordinary people do?

Ordinary people are not powerless here. We can support clear rules before crisis forces rushed rules. We can ask elected officials what standards they support for autonomous systems. We can encourage schools, cities, churches, hospitals, and event venues to create response plans. We can resist panic, but also resist denial. And we can reward companies that build safety into their products instead of treating safety like an optional upgrade.

Complexity cannot become camouflage. If a system is too complicated to explain after it harms someone, it may be too dangerous to deploy.

A useful test might be this: could the public understand who was responsible, what rules governed the system, what safeguards were in place, and how the harm would be investigated if something went wrong? If the answer is no, then we are not looking at responsible innovation. We are looking at trust without evidence.

None of that solves everything. Bad actors do not file paperwork. But good rules still matter. Seatbelts did not end car crashes. Building codes did not end fires. Background checks did not end violence. But each one made society less careless about predictable danger. That gives me some hope. We have done this before. We have looked at powerful tools, admitted the risk, and built habits, standards, and laws that made ordinary life safer.

So what do we do with this truth?

This is where I want to slow down and ask you, maybe across the table, while the coffee is still warm, a few questions:

• Should machines ever be allowed to make life‑and‑death decisions without a human in the loop?

·         If the technology exists and is cheap, what kind of control is realistic?

·         What should we require before any government, company, or contractor can deploy a system that uses AI to help apply force?

·         How do we protect innovation without pretending every innovation is harmless?

• And what does “security” even mean in a world where a single person can deploy a weapon that used to require a nation‑state?

These aren’t rhetorical questions. They’re the kind of questions a student of truth has to wrestle with.

Because truth isn’t just about what is. It’s about what could be, and what we choose to do about it.

A final thought over the last sip of coffee

The article ends with a simple statement: “We must decide whether a human should always be involved… or whether machines can be allowed to act alone” .

That is the crossroads we’re standing at. Not just whether machines can act alone, but whether we will have the wisdom to set limits before convenience, fear, and competition set them for us.

And the decisions we make now, or fail to make, will shape not just warfare, but everyday life, public safety, and the boundaries of human responsibility.

I don’t have all the answers. But I do believe we have more agency than fear sometimes tells us. We can keep asking the right questions. We can keep human dignity at the center. We can insist that power comes with responsibility. I’m not asking us to become afraid of every new tool. I’m asking us to become serious enough, and hopeful enough, to guide powerful tools before they guide us.

Maybe that is the Finding Truth posture here: patient enough not to panic, honest enough not to look away, hopeful enough to act, and humble enough to admit that technology needs conscience riding beside it.

If this kind of conversation resonates with you…

The first book in my Finding Truth series, A Student of Truth, is available now at major retailers, and at a discount on cletustaylor.com.

If you have thoughts, reactions, or ideas for future topics, I’d genuinely love to hear them. Email me anytime at hello@cletustaylor.com.

Come back next week. I’ll have good coffee and meaningful things to talk about.