AI Surveillance Is Making the Case for Privacy Stronger Than Ever

Everyone says they have nothing to hide until the machine has every message they ever sent.

In the age of artificial intelligence, that space is being crushed.

The problem is no longer simply that companies collect data. The more serious danger is that AI can join the scattered fragments of a life and turn them into something searchable, analysable and usable. A phone location, a bank payment, a search query, a photograph, a message timestamp, a medical worry, a workplace complaint, a political hesitation — each may seem trivial on its own. Combined at scale, they become a profile. Processed by machine, they become a judgment.

That judgment may be wrong. It may be partial, crude or absurdly framed. But once it is accepted by an employer, insurer, platform, bank, police force, immigration officer, advertiser or intelligence agency, its accuracy may matter less than its authority. The machine does not need to understand you in order to harm you. It only needs to classify you.

Meredith Whittaker, the president of Signal, has become one of the sharpest critics of this new order. Her critique carries weight because she is not an anti-technology romantic shouting from outside the walls. She spent years at Google, working inside the machinery of search, infrastructure, privacy, security and early artificial intelligence. She saw how data came to be treated as if it were a neutral mirror of reality.

Data is made. It is selected, structured, measured, cleaned, labelled, bought, sold and interpreted. It reflects the purposes of the systems that gather it. In the modern internet, those systems were built largely to answer the questions of advertisers: who is this person, what category do they belong to, what do they want, what are they likely to do, and how can they be reached?

That is why the privacy debate cannot be separated from the business model of the internet. Surveillance advertising did not merely produce irritatingly accurate adverts. It built a system in which human behaviour became raw material. It trained companies to extract intimacy, convert it into prediction and sell access to the predicted person. AI does not break from that history. It feeds on it.

The old internet collected. The new internet infers.

Much of the public debate about AI misses this. We are invited to stare at the spectacle: chatbots, synthetic images, automated essays, office work performed at speed. But beneath the spectacle sits a harder reality. The AI age depends on a small number of technology giants that control cloud infrastructure, app ecosystems, operating systems, advertising networks, data centres and the computational power required to train and deploy these models.

Imagine that every text message you have ever sent is placed in a database. Not the careful messages. All of them. The angry ones. The frightened ones. The stupid ones. The jokes that would curdle if repeated in public. The private anxieties. The messages to lovers, doctors, lawyers, friends, colleagues and family members. The things you said before you understood the issue. The things you said before you became a better person. The complaint about your boss. The confession of weakness. The political thought you tried out and abandoned.

Now place an AI system over that database. Let anyone with access search not only by name and date, but by theme, implication, mood and inferred motive. Find every lie. Find every insult. Find every mention of depression. Find every sexual reference. Find every contradiction. Find every private sentence that can be made ugly by stripping away context.

Almost nobody would accept that exposure. Almost nobody could survive it with dignity intact. That is not because everyone is secretly criminal. It is because everyone is human.

Privacy is the architecture of ordinary dignity. A person does not speak to his closest friend as he speaks to his oncologist. He does not speak to his lawyer as he speaks to his child. He does not speak to his spouse as he speaks to his employer. That is not deception. It is social life. A world without privacy is not a world of perfect honesty. It is a world of permanent performance.

Artificial intelligence threatens privacy not only by processing what is collected, but by creating new information through inference. It can generate summaries, risk scores, predictions, psychological sketches, behavioural categories and reputational signals. It can produce a story about who someone is, and that story can travel faster than explanation. It can be used to decide what a person sees, what he is offered, what he is denied, whether he is trusted, whether he is investigated, whether he is employable, insurable, admissible or suspect.

The cruelty is that much of this does not depend on meaningful consent. People are tracked by phones, apps, cookies, cameras, payment systems, workplace software, school platforms, smart devices and the basic machinery of modern life. The answer often given is that users agreed. But agreement has become a ritual, not a safeguard. The choice is usually between participation and exclusion.

Consent under those conditions is not freedom. It is paperwork.

Nor is this only a corporate danger. The state is moving into the same architecture. Governments have always wanted information. What is different now is that private companies have already built vast systems of surveillance, classification and prediction. The state does not always need to build its own machinery. It can demand access, subpoena records, purchase brokered data, contract with cloud providers, or embed commercial AI systems into policing, welfare, defence, intelligence and border control.

The merger of state appetite and corporate surveillance is the real frontier. It is one thing for a government to investigate a suspect under law. It is another for a society to permit the routine accumulation of intimate data about everyone and then hope that power will remain restrained.

Hope is not a constitutional principle.

The usual answer is safety. Terrorism, child abuse, violent threats, suicide, fraud and organised crime are invoked whenever officials demand broader access to private communications. These harms are real. Any serious defence of privacy must admit that there are cases where intervention is necessary and where the state would fail in its duty if it did nothing.

But that admission does not end the argument. It begins it.

The question is whether weakening privacy infrastructure for everyone actually prevents those harms, or whether it creates a general power that will later be used for other purposes. A backdoor built for the noble case remains a backdoor. A scanning system justified by the worst crimes will not necessarily stay confined to the worst crimes. A database created for safety can become a database for immigration control, political monitoring, labour discipline, social sorting or repression.

That is not fantasy. It is how power works. Powers granted in moments of fear rarely remain confined to the fears that produced them.

Encryption sits at the centre of that fight. Its critics call it a shield for criminals. Its defenders call it one of the last barriers protecting ordinary life from institutional overreach. The truth is that encryption can protect both the innocent and the guilty, as all rights can. Lawyers defend the guilty and the innocent. Medical confidentiality protects the admirable and the disgraceful. Due process slows the punishment of the dangerous as well as the harmless. That is not a defect in the rule of law. It is the price of preventing power from becoming arbitrary.

Signal matters because it represents a contrary design principle: collect less, know less, retain less, and make it technically difficult to surrender information the service does not possess. That does not make any platform perfect. Devices can be compromised. Users can make mistakes. Private tools can be misused. But the principle is essential. Privacy should not depend on corporate mercy, official restraint or the hope that nobody important takes an interest in you.

For many people, that hope is now the last refuge: keep your head down, say little, do not attract attention, and perhaps the machinery will pass you by. That is not liberty. It is conditional invisibility.

The deeper intellectual error of the AI age is the belief that data knows us better than we know ourselves. A data profile can record behaviour, but it cannot fully understand a life. It can measure, correlate and predict, but it cannot capture irony, remorse, fear, coercion, growth, ambiguity or the private distance between a thought and an action. It can tell a story about someone. It cannot finally say who that person is.

Yet more and more institutions behave as if it can.

That is the real privacy crisis. Not simply that secrets may be exposed, but that the right to author one’s own life is being transferred to systems built by others, for purposes most people never chose. The machine does not merely observe. It classifies, narrates and judges. It turns conduct into evidence and evidence into destiny.

A democratic society needs areas of life that are not constantly measured. It needs confidential conversations, private reading, secure journalism, protected legal advice, medical intimacy, political dissent, religious exploration, sexual privacy, family life and the ordinary right to be foolish without becoming a permanent file.

The question, then, is not whether we have something to hide. Everyone does. The question is whether a free society can survive once every private fragment of life can be collected, joined, searched and turned into a profile by institutions powerful enough to act on it.

Privacy is not the enemy of safety. It is one of the conditions of freedom. And in the AI age, what cannot be kept private will eventually be used by someone else to decide who we are.