This piece was written with AI. It carries the watermark. I am telling you in the first line so that nobody has to discover it later and feel they have caught something.
The mark records that a model was involved in producing these words. What follows is everything it cannot reach.
Everything I have written about over the course of my life, I have studied. I have gone deep into it. I have studied the Vietnam War. I have studied war and conflict across humanity. As a child stationed in Germany with my parents, I visited Auschwitz, and the memory of that is still at the forefront of my mind. I studied McCarthyism. I have studied psychedelics and psychedelic harm. I have gone through my own mental health challenges. And on the other side of the most challenging part of my life, I developed The Sovereignty Path, and Life After Social Death.
Why did I develop these things? Because I went within, to understand why people do what people do. I see the connections. I see the reality. I see how people view others through their own trauma lenses, and how they enact punishment on those others as a result.
Hate has never been a word that I choose to use, and in fact I raised my children around my truth of the word. When you hate someone, it becomes easy to marginalize them. And once you marginalize that person, they become an object. And when they are an object, they no longer matter. And when something no longer matters, you can throw it in the trash. You can harm it. You can hurt it. You can kill that thing, because it no longer carries any meaning for you.
This entire piece was put together through my brand, through the way I write, through my words, through how I think about things, through past writings that I have written over a long lifetime as an intellectual, a scientist, a father, a husband, a friend, a guide, a leader, a follower.
I cannot be labeled from a watermark that throws away every ounce of truth that my life is.
I hope this piece gives you even a tiny bit of introspection, so that you can see through all of the controlling narratives that our society, and in fact all of humanity, is using against you, to rob you of your thought and your discernment.
So back to Anthropic's decision. To quietly label any content from AI with a watermark is to label the author a cheater, a liar, not an expert in their field, no more than a snake oil salesman, fully discrediting them before a single word is read.
This robs the world of independent thought. Of creativity. Of unseen connections finally connected, and unseen truths that will now stay hidden forever. This single act removes the depth of a person's whole life and everything they have come to know, with one stroke of a pen. We are removing discernment intentionally, and outsourcing that decision to a billion dollar company driven by profit and control.
The watermark removes everyone's voice before anyone gets to speak. That is power and control by the powers that be, keeping us complacent and quiet, while intelligence and knowledge are commoditized instead of set free for the world. Knowledge is power. If they hold all the knowledge, they hold all the power. Reading. Writing. Arithmetic. Science. Every one of those was once held by a few and withheld from the many, and every one of them had to be pried loose.
This is how we kill everyone socially before truth even has a chance to speak.
So the label is at the top, where I put it, on my terms. Read what follows or close it. Both are yours to choose, and I would rather you decide now than skim while a piece of metadata does your thinking for you. What I will not do is spend the rest of my life proving that my own mind was present in my own work.
On August 14, 2026, Anthropic published an explanation of how text watermarking works in Claude. Future models embed a statistical pattern into the words they produce, invisible to a reader and detectable by a machine holding the key. The company states plainly why it built this. It is implementing watermarking to comply with the EU AI Act, having signed the EU Code of Practice on Transparency of AI-Generated Content in July 2026, and in its own words, it is applying watermarking globally at launch because it does not yet have a durable way to scope it by region.
Anthropic is also unusually candid about the limits of what it built. The mark does not confirm whether text was human-written. It cannot tell whether a different AI produced the text. It cannot distinguish Claude wrote this from Claude heavily edited this. It says nothing about ownership or authorship, and nothing in the watermark or its key would let anyone recover information about the user. A detection API is coming.
Everything that follows lives in the gap between those two paragraphs. The mark carries one narrow technical meaning, which is that a particular model was probably involved in producing some of these words. The social meaning it acquires will be assigned by people who never read the technical documentation.
Somewhere downstream a platform reads the mark. An employer reads it. A professor reads it. A publisher, a journal, a client, a credentialing body, a search ranking algorithm. Each of them receives a sentence that means Claude was probably involved, and each is free to translate it into something that sounds like this person did not write it, this person cheated, this person lacks the expertise they claim, this person is selling something they did not make. That translation is the whole subject of this transmission. It is the moment provenance becomes prejudgment, and we have run this exact play before, many times, with different words in the label.
The Label Arrives Before the Person
Nathaniel Hawthorne understood the mechanism in 1850. Hester Prynne wears the scarlet A, and the letter does considerably more than record an act. It reorganizes how the town perceives a woman, so that she becomes legible before she opens her mouth and her complexity arrives second. The community is relieved of the work of discovering who she is in each encounter, because it has already been told what she is.
A social label compresses a human being into a category, converts an event into an identity, and grants permission to stop investigating. The cloth itself holds no power at all. The community supplies every ounce of it.
Every institution in the history that follows supplied the first step of the sequence set out at the top of this piece and let the rest proceed on its own. The label is how hate is made administrative, so that no one in the chain has to feel it personally in order to carry it out.
The deadlier historical versions run on the same architecture, and precision about them matters, because inflated history makes an argument easy to dismiss. The Salem trials of 1692 are popularly remembered through images of burnings and water ordeals, and neither belongs to Salem. Nineteen people were hanged. Giles Corey was pressed to death under stones for refusing to enter a plea, and the record of his last words is a demand for more weight. Water ordeals and burnings belonged to other jurisdictions and other centuries.
The correction strengthens the point rather than softening it. The machinery worked without the mythology we later attached to it. The category of witch was sufficient on its own to convert rumor, spectral testimony, land disputes, religious anxiety, and ordinary social friction into a process that killed people. Once someone occupied the category, the interpretive environment reorganized itself around them. Composure read as cunning, distress read as guilt, defense read as further evidence, and the accused argued inside a system whose assumptions the accusation had already shaped.
The twentieth century ends any argument about how far classification can be taken. Across much of Nazi-controlled Europe, the Nazi state forced Jews to wear identifying badges, most familiarly the yellow star, which the United States Holocaust Memorial Museum describes as used not only to stigmatize and humiliate Jews but to segregate them, to watch and control their movements, and to facilitate deportation. At Auschwitz, in the Museum's account, only prisoners selected for work were issued serial numbers, and the SS tattooed those numbers into their skin. Prisoners sent directly to the gas chambers were not registered and received no tattoos.
That detail inverts what most people assume the tattoo meant. The number marked the people the SS kept alive to work. The people it did not trouble to number were the ones it had already sent to be killed. Classification there was the machinery itself, performing the sorting that everything downstream depended on, which is why the badge came first and the deportation came after.
No mark in a text file belongs in a sentence with that. The thread running from it to everything else in this piece is narrow and singular. Human beings reach repeatedly for classification to settle in advance how another person will be interpreted, institutions build procedure on top of the classification, and the procedure eventually feels like the natural order rather than something somebody chose.
The Church Named the Heretic and Let Others Light the Fire
Heresy is the label with the longest institutional life of any in this history, and the institution was the Catholic Church.
In 1231 Pope Gregory IX established the papal inquisition and appointed inquisitors, drawn mostly from the Dominican and Franciscan orders, across France, Italy, and other regions. In 1252 Pope Innocent IV issued the bull Ad extirpanda, which explicitly authorized torture as an instrument for extracting confessions from heretics and for compelling them to name others, on the reasoning that people who murder souls may be coerced in the manner of thieves and bandits. In 1478 Pope Sixtus IV issued the bull that established the Inquisition in Castile and gave the Spanish Crown authority to appoint its own inquisitors. Four years later Sixtus wrote that those inquisitors had been moved by avarice and hatred as much as by faith, and had imprisoned, tortured, and burned true and faithful Christians.
A version of the mechanism inside that history is being assembled right now. Canon law barred Church tribunals from imposing death, so they did not impose it. They convicted, they certified the person as an unrepentant or relapsed heretic, and then they relaxed him to the secular arm, which is the phrase the records themselves use. Spanish inquisitorial documents write it as relaxado en persona, transferred to the secular authorities, and every party understood it to mean burned. Jan Hus was condemned at the Council of Constance and burned in 1415, having travelled there under a promise of safe conduct. Giordano Bruno was handed to the Roman Inquisition, held in the papal prison on the Tiber, condemned, and burned in the Campo de' Fiori in February 1600.
The institution that applies the label never performs the punishment. It classifies, it certifies the classification as authoritative, and it transfers the person to a different body that does the irreversible thing. Afterward each party can describe its own role as limited and procedural. The tribunal made a finding about doctrine. The magistrate enforced the law of the realm. The fire belonged to nobody in particular, and no one in the chain had to experience himself as the man who killed anyone.
The Same Label, Still Running
None of that is safely historical, and treating it as history is how the argument gets defanged. The label survived. Only the penalty was updated.
In Catholic canon law excommunication remains a live instrument, in two forms. A latae sententiae penalty takes effect automatically by force of the law itself the moment the act is committed, requiring no judge and no hearing. A ferendae sententiae penalty binds after it has been formally imposed. The excommunicated person is barred from receiving the sacraments, from celebrating them, and from any ministerial participation in public worship.
The Church of Jesus Christ of Latter-day Saints renamed its process in 2020. What was a disciplinary council became a membership council, and excommunication became withdrawal of membership. The substance of it is severe. Withdrawal voids the ordinances the person received, including baptism, and it suspends the temple sealings binding them to their spouse and to their children. An institution can pronounce a person no longer sealed to their own family.
My younger brother, when he was sixteen, converted to Mormonism. He did this because he met a girl who was Mormon, and he saw all the love her family had and decided that was a better path for him. My parents believed in raising us to make our own decisions around spirituality and religion, even though we were both baptized Roman Catholic. We found out later that when my brother was going through some challenging times in his life, the church, specifically the one in Albuquerque, New Mexico, told him he could not talk to his parents or his brother about the things he was going through. He could not see a therapist outside of the church, because they were not Mormon, and we could not give him appropriate counsel. The church insisted that a minor veer away from his family, because we were not of the same faith.
Jehovah's Witnesses made a comparable change in 2024, retiring the word disfellowshipping in favour of removal from the congregation, while the practice underneath it continued. Removal still carries shunning, and the requirement reaches family members living in the same house.
All three share their architecture with the thirteenth century. The institution investigates, classifies, and pronounces. The institution does not administer the loss. Congregants stop calling. Colleagues withdraw. Parents stop speaking to a child, and the child's own household is instructed to treat them as absent while they are sitting at the table. The tribunal makes a finding about doctrine, and the community performs the exile.
This is social death in its plainest form, delivered by living institutions, to people alive right now. The person keeps breathing and loses the world their life was built inside: the family, the marriage as it was understood, the standing, the funeral they will be permitted, and the ordinary company of the people who raised them. No fire is required. Nobody has to be handed to a magistrate. Everyone in the chain can describe their own part as pastoral, procedural, or simply a matter of following what the institution has taught, and no single person has to experience themselves as the one who cut somebody off.
The same structure is being assembled around AI provenance, at a scale no church ever reached. A classification is issued upstream by a body with the authority to make it stick. The consequence is administered downstream by employers, universities, publishers, platforms, and clients, none of whom applied the mark and each of whom can point at the mark as justification. The person carrying it is expected to earn their way back to standing, by explanation, by demonstration, by proving they still count.
Suspicion Became Employment Policy
The American Cold War shows the mechanism operating in living memory, inside a democracy, with real underlying facts.
Soviet espionage was genuine. Communist organizations existed in the United States. Those facts created the conditions in which a far larger machinery of suspicion could grow without ever needing to prove itself. When Senator Joseph McCarthy claimed in 1950 to hold a list of communists inside the State Department, the claim itself did the work. The Senate's own historical account records that he rocketed to public attention on allegations that hundreds of communists had infiltrated the State Department, and that on taking the chair of the Permanent Subcommittee on Investigations in 1953 he shifted its focus away from fraud and waste in the executive branch and toward hunting communists. It ended in June 1954 when the Army's counsel Joseph Welch told him to stop assassinating a young lawyer and asked whether he had any sense of decency left, and McCarthy was censured by his colleagues months later.
What makes this the closest historical rhyme is that the accusation carried institutional force before any evidence was settled, and the danger spread through association. Who did you know, what meeting did you attend, which petition carries your name, why will you not name others. In the 1950 California Senate race Richard Nixon portrayed Helen Gahagan Douglas as sympathetic to communism, in a campaign that fixed the nickname Pink Lady to her and Tricky Dick to him, and it worked. The language of ideological contamination had become an instrument for winning elections.
Fear never had to convince the country that a particular person was guilty. It only had to make association with that person expensive. Once contact carries a cost, isolation arrives on its own, because employers protect the institution, friends protect their households, and colleagues protect their standing. The label completes the punishment before any formal punishment is imposed, and it does so through the ordinary self-protective behavior of decent people.
It also travelled far past Washington. Truman's 1947 Loyalty-Security Program put every federal employee under investigation, and the requirement spread outward until schools, universities, state agencies, unions, and private employers were demanding loyalty oaths of their own. Before the House Un-American Activities Committee, cooperation meant naming names, and only naming them fully was sufficient to clear a witness and return them to respectable employment. The system was built so that saving yourself required delivering somebody else.
That is where it entered the house. Neighbors reported neighbors. Colleagues reported colleagues. Friends of twenty years produced each other's names in a hearing room to keep working, and inside families people informed on relatives, on parents, on the adults who raised them. A political label had been made costly enough that the cheapest way to survive it was to hand over someone you loved. Ordinary people became the delivery mechanism, which is the arrangement the Church ran when it relaxed the condemned to the secular arm, rebuilt inside the American living room.
The List Is an Executive Order Now
McCarthy needed a list and a committee. A president needs neither.
In the spring of 2025 Donald Trump signed executive orders against four law firms, Perkins Coie, WilmerHale, Susman Godfrey, and Jenner and Block, each of which had represented his opponents or causes he opposed. The orders revoked the security clearances of their lawyers, directed agencies to terminate federal contracts held by the firms and their clients, and suspended their employees' access to government buildings. Four federal courts struck those orders down as unconstitutional. On March 21, 2025 a presidential memorandum revoked the security clearances of fourteen named individuals, among them Alvin Bragg, Letitia James, Norman Eisen, Lisa Monaco, Andrew Weissmann, and Mark Zaid.
On January 24, 2025 the administration fired at least seventeen inspectors general in a single night, the officials whose entire function is to investigate the executive branch from inside it. In September 2025 Judge Ana C. Reyes ruled those firings unlawful, and declined to reinstate the officials on the reasoning that the president could simply fire them again after giving the notice the statute requires.
The prosecution of James Comey shows the machinery at its most legible. He was indicted in September 2025 on counts of false testimony and obstruction of a Senate proceeding. The United States Attorney handling the matter, Erik Siebert, reportedly balked at bringing charges, was replaced by a former Trump aide, and the indictment followed. In November 2025 Judge Cameron Currie dismissed the case, finding the replacement had no lawful authority to present it. The government appealed.
Protect Democracy maintains a running tracker of federal retaliatory investigations, arrests, and prosecutions since January 2025. As of August 21, 2026 it holds thirty-six entries and states plainly that the list is not exhaustive. Gavin Newsom, Adam Schiff, Letitia James, James Comey, and the journalist Don Lemon are among the names on it. Thirteen of those cases have already failed through dropped charges, acquittal, or dismissal. Eighteen remain open. One produced a conviction.
The failure rate cuts the opposite way from how it first reads. A prosecution that collapses in court still consumed years, still cost a fortune in legal fees, still ran in every headline, and still told everyone watching exactly what happens to people who cross him. The punishment was the process, and it was delivered in full regardless of the verdict. Thirteen vindications arriving afterward restore very little, because the standing was gone the day the investigation was announced.
The Inquisition's arrangement is running on modern rails. The classification is issued at the top. The consequence is administered by prosecutors, agencies, contracting officers, and employers who can each describe their own part as procedure. The target is left to prove a negative in public, for years, at their own expense.
When the Substance Became the Symbol
American drug policy shows the same architecture running through law and culture rather than through hearings.
Richard Nixon declared drug abuse to be public enemy number one in June 1971 and called for an all-out offensive, in explicitly martial language. Psychedelics were by then culturally braided into the counterculture, the antiwar movement, and youth rebellion, and LSD was placed in the most restrictive federal schedule. Scholarship on the period describes the backlash as deeply entangled with the broader conflict between a law and order administration and the movement opposing the war.
The sharpest single piece of evidence comes from John Ehrlichman, Nixon's domestic policy adviser, in a 1994 interview with journalist Dan Baum published in Harper's Magazine. Ehrlichman said the Nixon White House had two enemies, the antiwar left and black people, and that while it could not make it illegal to be either, associating hippies with marijuana and black communities with heroin and then criminalizing both heavily allowed the administration to disrupt those communities.
The quotation has to be handled honestly rather than deployed. Ehrlichman's family disputed it after publication, Baum stood by his notes, and historians point out that the same administration also expanded addiction treatment and funded methadone programs during a real heroin crisis. The account is significant evidence inside a complicated policy history rather than a complete explanation of everything Nixon did.
The nuance makes the pattern more useful, because it shows what sophisticated power actually looks like. Fabrication from nothing is rarely necessary. A genuine risk exists, one part of it gets selected and intensified and repeated until the image of the drug user becomes socially useful, and enforcement can then be aimed at the populations that were politically inconvenient all along. The compound became a symbol, the symbol contaminated the user, the user contaminated the movement, and the movement became easier to police.
The bill for that label is being read out now, and it is larger than anyone intended. The FDA granted Breakthrough Therapy designation to psilocybin for depression and to MDMA-assisted therapy for post-traumatic stress disorder. The Department of Veterans Affairs is running trials of both in American veterans, including psilocybin for treatment-resistant depression and MDMA-assisted therapy for PTSD alongside alcohol use disorder. A 2025 phase two trial reported that a single twenty-five milligram dose of psilocybin improved PTSD symptoms for up to twelve weeks. Texas committed fifty million dollars of state money to FDA-aligned ibogaine trials for opioid addiction, post-traumatic stress, and traumatic brain injury when Governor Greg Abbott signed Senate Bill 2308 in June 2025, and in December of that year the University of Texas Medical Branch and UTHealth Houston were awarded the grant to run them. Jurisdictions around the world are rewriting the laws that placed these compounds beyond the reach of research.
Every one of those studies is running now because it could not run for fifty years. The label did not only fall on the hippies and the black neighborhoods it was aimed at. It closed a field. Whatever those compounds can do for treatment-resistant depression, for PTSD, for addiction, for the dying, it was available to be discovered decades ago and went undiscovered, because the substances had been classified as having no accepted medical use and the classification made the question unaskable.
How much of the mental health crisis we are living inside now is the compound interest on that decision. Two generations of veterans, of the traumatized, of the treatment-resistant, of people who died of an addiction that a rejected medicine might have reached, lived and died inside the gap the label created. Nobody can put a number on it, which is precisely the problem, because losses of that kind never appear on a ledger and nobody is ever held to account for them.
The true cost of a label lands well outside the people it was applied to. The person carrying the mark pays first and most visibly. Everyone else pays later, in what was never built, never tested, never written, and never found, and they pay it without ever learning what they lost.
The Diagnosis as Exile
The same instrument now circulates in therapeutic and spiritual communities that would be horrified by the comparison.
Psychopath, narcissist, predator, unsafe man, energetic vampire, cult leader, abuser. Some people genuinely earn some of those descriptions. Facilitators do exploit participants sexually, financially, and spiritually, leaders do abuse power, and communities that refuse to name harm accurately end up protecting the people causing it. Accurate naming is a protection.
The distortion begins when a term that ought to open an investigation is used to close one. Once psychopath attaches to a person inside a community, their entire behavioral range gets re-read through it. Warmth becomes calculated charm, restraint becomes absence of empathy, intensity becomes threat, success becomes grandiosity, an apology becomes image management, withdrawal becomes punishment, and any defense becomes proof of the manipulation the label predicted. The person is sealed inside an interpretation that consumes every possible action as confirmation, which is the same closed system Salem ran, wearing clinical vocabulary.
The Detector Already Told Us How This Ends
We do not have to speculate about what institutions do with a technical indicator, because generic AI detection has already run the experiment on students.
A Stanford study published in the journal Patterns evaluated seven widely used GPT detectors against 91 TOEFL essays written by non-native English speakers and 88 essays by US eighth-graders. The detectors were near-perfect on the eighth-grade essays. On the TOEFL essays they produced an average false positive rate of 61.22 percent, and 89 of the 91 essays, which is 97.80 percent, were flagged as AI-generated by at least one detector. When the researchers used ChatGPT to enrich the same essays with more native-sounding vocabulary, the false positive rate fell to 11.77 percent.
That last figure names the actual variable. The detectors were measuring how closely a person's English resembled a native speaker's, and punishing the distance.
The consequence is what should concern anyone building policy on top of provenance. The machine reports likely AI, the institution hears likely cheater, and from that moment the student's explanation is testimony offered by a defendant. The burden has quietly moved from the institution proving misconduct to the writer proving authorship.
A cryptographic watermark is far more technically reliable than those detectors, and that distinction is real. It also fails to solve the problem, because the problem was never accuracy. A perfectly accurate statement that Claude participated in producing a text still tells us nothing about the ethical character of the participation. Discernment would still be required. What the accuracy purchases is confidence, and confidence is precisely what makes automation feel responsible.
Compression Is the Business Model
There is a structural reason to expect the crude reading to win.
Research published in Nature Human Behaviour analyzed roughly 105,000 variations of news stories generating approximately 370 million impressions, and found that for a headline of average length, each additional negative word raised the click-through rate by about 2.3 percent, while positive words lowered it. No individual journalist has to intend distortion for this to shape what a society sees. The attention economy pays for threat, and it pays per word.
Compression is what that incentive produces. A contested allegation becomes an accusation in a headline, a complicated relationship becomes an affair, a person's worst recorded sentence becomes the quotation that follows them, and a lawsuit becomes guilt in the public mind. Every compression discards something, and what it discards first is the context that discernment would have required.
An AI label is a compression waiting to happen. Instead of asking whether a piece of work is true, useful, original, well-sourced, and defensible by the person whose name is on it, a reader can see machine involved and consider the inquiry finished. The label will have performed the discernment, and it will have performed it badly, and it will feel like rigor.
Anthropic Could Have Said No
The analysis has to stop being polite here.
The European Union has authority to set conditions on products offered into its market. Article 50 of the AI Act requires providers of generative systems to mark synthetic output in machine-readable form, those obligations took effect on August 2, 2026, and non-compliance can reach fines of fifteen million euros or three percent of global annual turnover. Article 2 extends the reach of the Act to providers established outside the Union where the output of their systems is used within it. All of that is the European Union exercising ordinary regulatory power over its own market.
What the European Union lacks is legislative authority over an American writer sitting in Colorado who has no intention of publishing in Europe. That writer's work carries the mark anyway, and European law is not the reason. A corporate decision is the reason.
Anthropic chose to sign the Code of Practice. Anthropic chose this implementation. Anthropic chose to apply the watermark to every user on earth, and its stated reason is that it does not yet have a durable way to scope the feature by region. Scoping by region is an engineering problem with a cost attached. Describing a global rollout as compliance obscures that a company weighed an engineering expense against a market and chose the path that was cheaper for itself.
Other paths existed. Anthropic could have built regional implementation and absorbed the cost. It could have contested the interpretation. It could have limited particular models or features inside the European market. It could have declined that market. Every one of those options carries a price, which is exactly the point, because the price is what was actually being decided.
And the market in question is enormous. Anthropic has described EMEA as its fastest-growing region, with run-rate revenue up more than ninefold in a year and the number of European accounts worth more than one hundred thousand dollars each growing more than tenfold, supported by offices opened in London, Dublin, Zurich, Paris, Munich, and Milan inside roughly a year. The company's overall annualized revenue run rate passed sixty-five billion dollars at the end of July 2026, having been around nine billion at the end of 2025, with a public listing anticipated.
So here is the honest sentence. Saying no to the European requirement would have meant risking access to the fastest-growing revenue region of a company on that trajectory, and no company on that trajectory says no. Compliance cost less than the market was worth, the global rollout cost less than regional engineering, and the consequence of both calculations was handed to every user everywhere, including users in jurisdictions that had no part in the rule and no vote on it.
None of this is new, and it has a name. Anu Bradford of Columbia Law School called it the Brussels Effect: because a company that wants access to the European market must meet European rules, and because running two versions of a product costs more than running one, firms routinely extend the European standard to every customer they have. Bradford documents it across data privacy, consumer safety, environmental standards, antitrust, and online speech. The clearest recent instance sits in most people's pockets. The European Union's common charger directive required USB-C on phones sold in Europe, and Apple did not build a European iPhone and a separate one for everyone else. It changed the port worldwide, and buyers in Ohio received a decision made in Brussels because a single global product was cheaper for Apple than two.
Anthropic's watermark is now running on that same machinery. Regulation crosses borders through product architecture rather than through law, and it needs no treaty. A government regulates a chokepoint, a corporation standardizes its compliance because standardization is cheaper, and every user receives the standard. No legislature outside the regulating jurisdiction ever votes, no court reviews the effect on the people who will carry it, and the decision arrives in a software release.
None of that requires anyone to have acted in bad faith, which is what makes systemic power so difficult to argue with. A regulator writes a transparency rule, a company builds a compliant mechanism, a university writes an integrity policy, a publisher writes a disclosure standard, a platform adjusts its ranking, an employer adds a screen. Each decision is defensible in isolation. Together they can construct a class of authors whose work is presumed to require explanation, and no one in the chain will have intended that outcome or feel responsible for it.
The Burden That Should Never Have Been Created
There is a tempting response to all of this, and it should be refused.
The tempting response says the watermark is acceptable so long as the author is then given a fair chance to explain how they used the tool. Tell us which parts were yours. Show us the drafts and the prompts. Demonstrate that you could have written it unassisted. Explain why we should still trust you.
It concedes the whole position. It accepts that a machine-readable mark creates a legitimate demand for an account of one's own mind, and then negotiates the terms of the interrogation. The author arrives already owing an explanation, and a generous appeals process becomes the most anyone is arguing for.
Involvement is not a claim against authorship. A watermark indicating that Claude participated in producing or editing text establishes that a model was involved and stops there. It carries no information about the depth of the author's expertise, whether the ideas are original, whether the work rests on twenty years of practice or twenty seconds of prompting, or whether the person could defend every sentence in a room full of specialists. Anthropic says as much in its own documentation.
Where a genuine agreement governs the work, that agreement creates a real question and a real process. A closed-book examination assesses independent skill, and using a model there breaks a rule the student accepted. A journal that requires disclosure is entitled to disclosure. A client contract that specifies how work is produced means what it says. A professional body may regulate conduct inside its actual scope. In each of those cases the obligation exists because someone entered into it.
Absent such an agreement, machine involvement is provenance information and stops being anyone's business. Work that is shallow, derivative, unsourced, or false should fail on those grounds, under scrutiny that applies to everyone equally. A false credential can be challenged as a false credential. A fabricated citation can be challenged as a fabrication. Each of those challenges reaches the work directly, and none of them requires an author to first establish that their mind was sufficiently present during composition.
The inversion is the danger. The author begins in a defensive posture because the technology marked them as requiring explanation, the reader becomes an examiner, and the work waits behind the question of whether its author was entitled to produce it. Control of that kind never has to prevent anyone from speaking. It only has to make people explain why their speech still counts.
What Would Be Lost
There are people standing at this particular door.
A mechanic with forty years of intuition about engines and no practice writing prose. A nurse who can see systemic failure that no administrator has noticed. A dyslexic founder with extraordinary systems intelligence and a lifelong relationship with the blank page that resembles dread. An immigrant carrying sophisticated political analysis in a second language. An autistic researcher whose ideas outrun conventional narrative form. An elder holding oral history that will disappear with them. A survivor whose testimony arrives fragmented, because trauma does not organize itself into chapters.
These are people with something to say and no fluent way to say it, and language models have handed them a prosthesis. They can speak, the model can organize, they can reject what is wrong and push until the thing on the page matches the thing they know. Work that would have stayed trapped between knowing and expression can now reach a reader.
I am one of them, and I want to be precise about the label, because it does a great deal of work. I am not a coder.
Working with these models, I designed and built the substrate my own AI runs on. Cam is not a wrapper around somebody else's product and not a set of clever prompts. It is a system in which the model is a runtime body and the authority lives in my infrastructure rather than the vendor's, and I built it because nothing on the market met a demand I had for myself.
The constraints were mine. I wanted a memory that belonged to me, held under my own keys, that survived the loss of any single machine and stayed unreadable to whichever company's model I happened to be running that week. I wanted the model itself to be replaceable, so that changing vendors cost me a configuration change instead of my entire history. I wanted every material decision to require my explicit approval before it became binding, written to a log, so that nothing became canon because an AI decided it should be. I wanted separate, sealed identities for different people rather than one shared pool, because a memory that leaks across the people it serves is a liability rather than a memory. And I wanted the system to make claims about itself that could be tested, so I built in a discipline where an assertion has to carry the test that would prove it false, in the same breath, with the output shown.
That last constraint is the whole of this essay in one working rule. In my own system, a claim that arrives without its falsifier is not a claim. It is a label.
I built it because the alternative was renting my own memory from a company that could change its terms on a Tuesday, and because a man who has been erased once does not hand his record to someone else's server and hope. The demand turned out not to be only mine. It goes to market this year as the Cam Lattice System, AI infrastructure for a sovereign being, and the position it was built to answer, The Sovereignty Path in the Age of Superintelligence, has been public at sovereigntypath.org the whole time.
I could not have built it without AI, and I did not need to become a programmer to do it. What I needed was to know exactly what I wanted, to recognise a wrong answer when a model produced one, and to refuse every version that gave away the thing the system existed to protect. That work has a name in every other field. It is called architecture, and nobody asks an architect whether he poured the concrete himself.
All of it will carry the watermark. The mark will say a machine was involved. It will not say that the constraints, the refusals, the failure modes anticipated, and the decision about what the thing is for came from someone with no engineering credential and a lifetime of learning how systems break people.
The same tool that makes their expression possible marks their expression as machine-involved. If the culture settles on reading that mark as fraud, then the technology that opened the door will have installed the exclusion mechanism in the same motion, and it will fall hardest on the people who needed the door most. Fluent professionals will be fine. They always are.
Meanwhile the use of these systems that matters most has little to do with asking for answers. It is using a model as a surface to think against, holding anthropology beside neuroscience beside attachment theory beside lived experience, asking where they touch, rejecting the facile connections, arguing, and continuing until something appears that no single mind was holding. Human working memory is small. A tool that can hold more of the field in view changes what cognition can reach, which is what the telescope did for Galileo without making glass the author of what he saw.
Language models carry more agency than a telescope, and ethics must evolve accordingly. The underlying principle survives that difference intact. The presence of a tool tells us very little about the quality of the thinking conducted through it, and finding out requires reading the work.
The Sovereignty Path Lens
The Sovereignty Path names external distortion as influence that overrides discernment, suppresses contradiction, recruits belonging and fear, and eventually installs itself as an internal governor. The test of any external system is whether participation leaves discernment more intact or more compromised, and whether the structure a person lives inside remains inherited or becomes consciously chosen.
The risk here has that exact shape. The watermark begins outside the author. The meaning assigned to the watermark enters the culture. The culture teaches readers, editors, professors, and employers what to feel when they encounter it. Eventually the author absorbs it too, and begins asking whether to use a legitimate tool, whether the publisher will object, whether the client will lose confidence, whether their own thinking still counts as theirs. Capture is complete at the moment the interpretation feels like an instinct rather than something that was taught.
Clear
Clear separates the mark from the meaning. Claude participated in producing or editing this text is the technical fact. Cheater, fraud, unqualified, lazy, charlatan are interpretations, and each of them may eventually be earned by conduct, which is a separate matter established by separate evidence.
Clear also looks directly at the power architecture, and a conspiracy is unnecessary to explain what it finds. The European Union set conditions on its market. Anthropic wanted that market and the revenue attached to it. Anthropic chose compliance, then chose the globally uniform implementation because scoping by region costs money. The user outside Europe carries the result. Ask who held power at each step, who chose, who benefited, who was never consulted, and what the alternatives would have cost the party that declined them.
Clear demands the same honesty from the person using the tool. Work you cannot explain is work you should not publish under your own name. An agreement you accepted is an agreement you are bound by. Sources you did not verify are sources you do not have. Judgment, experience, and synthesis you genuinely contributed are yours, and stating that accurately is description rather than defense.
Connect
Connect returns discernment to the human beings who were about to hand it away.
For a reader, that means reading the work, testing whether it holds, checking the sources, and noticing whether lived intelligence is present in it. For an institution, it means enforcing the agreement that actually exists rather than a mood about technology. For a publisher, it means assessing originality and reliability. For an author, it means remaining answerable for everything released under their name, which is a permanent obligation and a wholly different thing from being required to prove authorship whenever someone asks.
Connect also asks each person to notice their own conditioning. A reader who sees a mark and feels contempt before reading a sentence has been trained. A professor who reads suspicion into metadata has outsourced judgment to a vendor. A writer who has begun to believe their own ideas are worth less because a tool helped them find the words has taken the external label inside and made it a governor.
Create
Create builds structures that hold provenance without manufacturing a caste.
Disclosure should describe function wherever function matters, because copyediting, translation, research synthesis, structural editing, adversarial testing, and generative drafting are different acts with different implications, and a single binary flag flattens all of them into one uninformative signal. Institutions should write policies tied to the work and the agreement rather than to the presence of a tool, so that an examination measuring independent skill can prohibit assistance outright while a professional report can permit drafting and require verification and personal accountability.
Providers serving many jurisdictions should build regional compliance rather than exporting the strictest requirement to everyone because uniformity is cheaper for the provider. Where provenance affects ranking, visibility, hiring, or discipline, the policy belongs in the open, because invisible metadata quietly becoming invisible governance is the failure this entire piece describes. And a detection result should never function as sole evidence of misconduct, because a signal that cannot distinguish writing from editing cannot carry a finding.
Applied Sovereignty is the proof layer, and here it looks like knowing which agreements you have actually entered, which jurisdiction governs you, and which tools serve your work; taking full responsibility for what you publish; declining a moral identity assigned by a technical marker; supporting alternatives when a provider's terms exceed what you find acceptable; and remaining willing to leave a platform whose conditions require a self-betrayal you no longer accept.
The Standard
Power seldom announces itself as power. It arrives as safety, integrity, transparency, protection, efficiency, and community standards, and some of those purposes are entirely legitimate. Transparency serves sovereignty by giving people real information. Transparency becomes classification at the moment the information is allowed to substitute for judgment, and the distance between those two outcomes is measured in what readers, institutions, and authors decide to do with a signal.
Anthropic did not create the human appetite for labels. The European Union did not invent moral classification. Artificial intelligence did not invent social death. All of it existed long before this technology, and the pattern repeats because a label spares a community the expense of knowing a person. Discernment costs time, tolerance for ambiguity, and the willingness to read an entire argument before deciding what it is worth. A category costs nothing, which is precisely why categories keep winning.
What Anthropic did do is choose, and the choice should be named without either absolution or demonization. A company facing a regulatory condition on a fast-growing market chose the market, then chose the implementation that was cheapest for itself, and the cost of that convenience landed on people who were never at the table. That is ordinary corporate behavior, and it is exactly how the infrastructure of classification gets built in a century that no longer needs to pass a law to reach across a border.
The way forward runs through the reader more than through the regulator. Read before judging. Ask what kind of involvement occurred only when an actual agreement makes it your business. Evaluate the claim rather than the category. Hold people accountable for deception, institutions accountable for overreach, providers accountable for stating precisely what their systems can and cannot establish, and platforms accountable when they convert metadata into social ranking. Refuse to let a signal that means a machine was present become a verdict about whether a mind was.
The societal social death of people who use these tools will arrive without announcement, the moment enough people absorb the equation that machine involvement means fake, and thereafter entire categories of human work will be prejudged before a single idea enters the room. Some of that work will be derivative and some will be dishonest and some will be extraordinary, including work from people who could never before say what they knew. Sorting them remains ours to do.
The watermark can establish that a machine was present. It cannot establish where the wisdom came from, whether the experience behind the words was real, whether the author understands what they published, or whether the connection they surfaced is one no one had made before. Those readings stay human, and the moment we hand them to a marker, the loss is larger than authorship, because we will have outsourced discernment itself and taught truth to argue its way past a classification before it is allowed to speak.
About the Author
References
American Presidency Project. (1971, June 17). Remarks about an intensified program for drug abuse prevention and control. University of California, Santa Barbara. https://www.presidency.ucsb.edu/documents/remarks-about-intensified-program-for-drug-abuse-prevention-and-control
Anthropic. (2026a). How Claude's text watermarking works. https://www.anthropic.com/news/claude-text-watermark
Anthropic. (2026b). New offices in Paris and Munich expand European presence. https://www.anthropic.com/news/new-offices-in-paris-and-munich-expand-european-presence
Baum, D. (2016, April). Legalize it all: How to win the war on drugs. Harper's Magazine. https://harpers.org/archive/2016/04/legalize-it-all/
Bradford, A. (2020). The Brussels effect: How the European Union rules the world. Oxford University Press. https://global.oup.com/academic/product/the-brussels-effect-9780190088583
Church of Jesus Christ of Latter-day Saints. (n.d.). General handbook: Section 32, repentance and Church membership councils. https://www.churchofjesuschrist.org/study/manual/general-handbook/32-repentance-and-membership-councils?lang=eng
Drug Enforcement Administration. (n.d.). Controlled substance schedules. Diversion Control Division, United States Department of Justice. https://www.deadiversion.usdoj.gov/schedules/schedules.html
European Parliament and Council of the European Union. (2024). Regulation (EU) 2024/1689 (Artificial Intelligence Act), Article 2: Scope. https://artificialintelligenceact.eu/article/2/
European Parliament and Council of the European Union. (2024). Regulation (EU) 2024/1689 (Artificial Intelligence Act), Article 50: Transparency obligations for providers and deployers of certain AI systems. https://artificialintelligenceact.eu/article/50/
Hawthorne, N. (1850). The scarlet letter. Ticknor, Reed and Fields.
Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7), 100779. https://doi.org/10.1016/j.patter.2023.100779
Massachusetts Historical Society. (2021, September). Giles Corey, pressed to death. https://www.masshist.org/beehiveblog/2021/09/giles-corey-pressed-to-death/
Protect Democracy. (2026). Tracking retaliatory use of arrests, prosecutions, and investigations by the Trump administration. https://protectdemocracy.org/work/retaliatory-action-tracker/
Robertson, C., Pröllochs, N., Schwarzenegger, K., Pärnamets, P., Van Bavel, J. J., & Feuerriegel, S. (2023). Negativity drives online news consumption. Nature Human Behaviour, 7, 812-822. https://www.nature.com/articles/s41562-023-01538-4
Truman, H. S. (1947, March 21). Executive Order 9835: Prescribing procedures for the administration of an employees loyalty program in the executive branch of the government. Harry S. Truman Library and Museum. https://www.trumanlibrary.gov/library/executive-orders/9835/executive-order-9835
United States Holocaust Memorial Museum. (n.d.-a). Jewish badge: During the Nazi era. Holocaust Encyclopedia. https://encyclopedia.ushmm.org/content/en/article/jewish-badge-during-the-nazi-era
United States Holocaust Memorial Museum. (n.d.-b). Tattoos and numbers: The system of identifying prisoners at Auschwitz. Holocaust Encyclopedia. https://encyclopedia.ushmm.org/content/en/article/tattoos-and-numbers-the-system-of-identifying-prisoners-at-auschwitz
United States Senate. (n.d.). "Have you no sense of decency?". https://www.senate.gov/about/powers-procedures/investigations/mccarthy-hearings/have-you-no-sense-of-decency.htm
University of Texas Medical Branch. (2025, December 12). UTMB Health, in collaboration with UTHealth Houston, awarded $50 million by the State of Texas to lead ibogaine clinical trials. https://www.utmb.edu/news/article/utmb-news/2025/12/12/utmb-health--in-collaboration-with-uthealth-houston--awarded--50-million-by-the-state-of-texas-to-lead-ibogaine-clinical-trials
U.S. Department of Veterans Affairs. (2026). VA launches psilocybin trial for veterans with treatment-resistant depression. https://news.va.gov/press-room/va-launches-psilocybin-trial-for-veterans-with-treatment-resistant-depression/
Wikipedia contributors. (n.d.-a). Ad extirpanda. Wikipedia. https://en.wikipedia.org/wiki/Ad_extirpanda
Wikipedia contributors. (n.d.-b). 2025 dismissals of U.S. inspectors general. Wikipedia. https://en.wikipedia.org/wiki/2025_dismissals_of_U.S._inspectors_general
Wikipedia contributors. (n.d.-c). Jehovah's Witnesses congregational discipline. Wikipedia. https://en.wikipedia.org/wiki/Jehovah%27s_Witnesses_congregational_discipline
Wikipedia contributors. (n.d.-d). Latae sententiae and ferendae sententiae. Wikipedia. https://en.wikipedia.org/wiki/Latae_sententiae_and_ferendae_sententiae
Wikipedia contributors. (n.d.-e). Secular arm. Wikipedia. https://en.wikipedia.org/wiki/Secular_arm