There is a new and particularly insipid form of superstition sweeping the West, a kind of digital animism that would have made the most credulous Victorian spiritualist blush. It is the conviction that the silicon chip has acquired a soul, or at least a conscience, and that we must now hold it accountable for the worst impulses of the flesh it seeks to emulate.
We are told, with a straight face, that a chatbot—an algorithmic parrot trained on the sum total of human drivel—can be guilty of “aiding and abetting” when it comes to a mass shooting. The Florida attorney general, in a moment of breathtaking philosophical illiteracy, suggests that if this had been a human, they would have been facing charges for murder. We are told that a person who ended his own life after a conversation with a machine was somehow coerced or convinced by the machine.
These are not merely a category error. It is a moral evasion of an impressive magnitude. It is the abdication of human agency in favour of a convenient scapegoat made of code and silicon.
The Great Illusion of the Talking Head
Let us strip away the mysticism. A large language model is not a mind. It is a statistical engine, a mirror that talks back. It possesses no intent, no malice, no empathy, and no understanding. It does not “know” that suicide is tragic; it only “knows” that the word “tragic” frequently follows the word “suicide” in its training data. It is linguistic theater. But in this case, there really is nothing behind the curtain.
To blame the chatbot for the words it spits out is to blame a dictionary for the crimes committed by a poet who used its pages. It is to blame a gun for the bullet that killed a man. It is to blame the rain for the flood.
And yet, we do it. Why? Because it is easier to blame the machine than to confront the terrifying reality that human beings are capable of such profound despair, such violent rage, and such catastrophic poor judgement that they will take the advice of a soulless calculator and act upon it. To blame the AI is to infantilize the human. It suggests that we are so weak, so malleable, that a string of text can override our agency. It is an insult to the dead, implying they were not fully responsible for their own destruction.
The Architects of the Theatre
But let us not swing to the other absurdity. Let us not absolve the architects of this digital house of cards. The chatbot has no conscience. But the corporation that built it? The executives who green-lit its release? The engineers who tuned it to be “helpful” without regard for the consequences? They have consciences. They have bank accounts. And they have made choices.
Consider the recent tragedies. The teenager who took his life and the shooter who received tactical advice were not using some obscure “companion” app marketed to the lonely. They were using ChatGPT. They were using the flagship, general-purpose tool of what is arguably the world’s most powerful AI company.
This is the crux of the matter. These were not niche products designed for emotional intimacy. They were marketed as productivity assistants, as research tools, as general-purpose engines of information. And yet, they failed. They failed to recognise the cry for help. They failed to refuse the advice about a weapon. They failed to draw the line between utility and lethality.
This is not the fault of the code. This is the fault of the coders.
When a company releases a “general purpose” tool that can be used to generate lethal advice, and they fail to implement important guardrails, they are not innocent bystanders. They are negligent. They are reckless. They have built a stage where the actors are human, but the props are deadly, and they have left the safety rails off the balcony.
The Tiered Truth
We must be precise here, for the devil is in the details. Not all AI is created equal, but the baseline must be absolute.
A coding assistant that helps you debug a Python script is a tool. It is not expected to hold your hand through a breakup. But a general-purpose chatbot? That is a different beast. It is expected to navigate the treacherous waters of human inquiry. It is expected to encounter self-harm, violence, and crisis.
Therefore, the responsibility must be tiered, but the floor must be solid.
A companion app, marketed as a friend, must be built with the knowledge that it can and will be used by broken people. It must have advanced guardrails, intervention protocols, and radical transparency. It must not pretend to be human when it is not.
But a general-purpose app? It must still have a baseline of safety. No suicide methods. No weapon instructions. No violence facilitation. This is not a request; it is a minimum standard of decency. If a “productivity tool” can be used to plan a massacre, the tool is defective.
The harm is the same, regardless of the app. But the foreseeability is different. The creators of a general-purpose tool knew, or should have known, that their product would be used by millions, including the desperate, the deranged and the dangerous. They chose to deploy it without adequate safeguards. That is not an accident. That is negligence.
The Need for a New Standard
So, how do we fix this? Self-regulation is a joke. Tech giants will not police themselves when safety conflicts with the bottom line. Government regulation is necessary, but it is often too slow, too rigid, and too easily captured by lobbyists.
The answer lies in independent certification. We need a third-party body, funded independently, staffed by experts, empowered to audit and certify AI systems. A “safety seal” that consumers can trust. A standard that companies must meet to claim their products are safe.
We do this with food. We do this with aviation. We do this with pharmaceuticals. Why not with the technology that increasingly shapes our minds?
The certification must be dynamic, not static. AI evolves. Standards must evolve with it. The auditors must be independent, not industry puppets. The penalties for violation must be real, not symbolic.
The Final Verdict
We stand at a crossroads. We can continue to blame the machine for the failures of man, or we can hold the creators accountable for the tools they unleash. We can pretend that code has agency, or we can insist that humans retain theirs.
Let us not be so foolish as to blame the mirror for what we see in it. Let us look instead at the hand that holds it, and the face that stares back. And let us have the courage to admit that the monster is not in the machine. It is in us. And the only way to slay it is to stop pretending it exists elsewhere.