Who Died and Made you King?

Who Died and Made you King?

Impolite or not, I think we need to start asking it of the companies building frontier artificial intelligence.

Who gave a handful of private technology companies the authority to determine the boundaries of human intellectual inquiry?

More importantly, what happens when their commercial interests, their definitions of credibility, and their decisions about acceptable information begin influencing what the rest of humanity gets to know?

These aren't hypothetical concerns about some distant future.

The systems are being built now. The policies are being written now. And our dependence on AI as an interface to knowledge is growing now.

We need to ask the questions before that dependence becomes the architecture of our intellectual lives.

What changed at Anthropic?

On October 8, 2026, Anthropic announced updates to the usage policy governing Claude, its AI assistant. The revised policy takes effect November 12.

Among the changes is a more explicit prohibition against using Claude to manipulate the sources that search engines and AI systems rely on for answers.

The policy specifically targets deceptive networks of websites that appear independent but are actually controlled by the same source and designed to corroborate the same claims.

SEO and AI search expert Lily Ray drew attention to the changes, highlighting reporting by Hanns Kronenberg and Matt G. Southern.

 

There is good reason to pay attention.

We've already seen what happens when AI-generated content floods the internet: fabricated expertise, fake reviews, synthetic consensus, and enormous volumes of information that offer little original value.

Why should we spend our time reading something nobody bothered to think about?

Anthropic says it has observed commercial organizations and influence operations using Claude to operate fake accounts and fabricated news sites. Its updated policy consolidates prohibitions that previously appeared in several different sections.

That's a legitimate problem. Deliberately manufacturing the appearance of independent authority corrupts the information environment.

But there's an important distinction.

Protecting information from deception is not the same thing as determining which knowledge deserves legitimacy.

Anthropic's published policy prohibits the former. It does not establish a general prohibition on independent thinking, intellectual synthesis, or challenging experts.

My concern is what happens as these systems become more powerful and the companies controlling them assume an increasingly influential role in evaluating information.

Because a rule designed to protect information quality can have consequences well beyond its original purpose.

And that's where this gets interesting.

What if the experts are wrong?

Around the same time, researcher Ethan Mollick was discussing another remarkable development.

He described using GPT-6 Pro to review academic papers and finding that the model could identify substantive weaknesses, suggest additional analyses, and recognize strengths rather than simply generating a laundry list of minor criticisms.

This doesn't mean the model's criticisms are automatically correct.

It means that AI may become an increasingly powerful instrument for examining established research—including work that has already passed through conventional academic review.

Think about the implications.

The very technology capable of identifying weaknesses in established expertise is also becoming a technology through which people access that expertise.

So whose judgment takes precedence?

The expert? The model? The institution? The company that determines how the model evaluates sources?

And what happens when any of them are wrong?

We already know that established scientific knowledge is fallible.

In 2015, the Open Science Collaboration attempted to replicate 100 published psychology studies. While 97% of the original studies reported statistically significant findings, only 36% of the replications produced statistically significant results.

That does not prove the remaining studies were false. It demonstrates why published findings require continued examination.

Expertise matters. Evidence matters. Methodology matters.

But expertise is not immunity from being questioned.

In fact, the capacity to challenge previous conclusions is one of the foundations of scientific progress.

So why would we want AI systems that might increasingly privilege established authority over potentially valuable challenges to it?

Isn't this how humans learn in the first place?

Every expert was once somebody learning from somebody else.

We read. We listen. We observe. We question.

We absorb what others have discovered, connect ideas, identify contradictions, and occasionally reach conclusions nobody else has considered.

That's how knowledge develops.

It's how disciplines evolve.

And it's how people outside established institutions sometimes make meaningful contributions to fields they were never officially invited to enter.

The originality of a thought doesn't necessarily come from inventing every piece of information that contributed to it.

Sometimes originality lies in the connections.

Consider the questions behind this very article.

Lily Ray raised an issue about Anthropic's treatment of deceptive information sources.

Ethan Mollick illustrated AI's growing ability to critique established research.

I connected those developments and began asking what they might mean for intellectual authority, commercial power, and the democratization of intelligence.

That's not manufactured corroboration.

That's human synthesis.

And it's precisely the kind of thinking AI ought to help more people engage in.

The opportunity was never simply to give people faster answers.

It was to expand the number of people capable of asking better questions.

Commercial advantage: Who benefits?

This is where the philosophical argument becomes a structural business question.

If AI systems increasingly determine which sources are credible, which institutions are recognized, and which businesses appear in responses, they can also influence who becomes visible, trusted, and commercially valuable.

Think about the difference between traditional search and an AI-generated answer.

A search engine typically presents multiple results for examination. An AI assistant can compress those sources into a single synthesized response, sometimes without requiring the user to visit any original source.

That makes the system's decisions about retrieval, selection, citation, and presentation extraordinarily consequential.

What gets included?

What gets left out?

Which sources are treated as independent confirmation?

Which institutions are considered authoritative?

Which businesses become the recommended choices?

And what about the independent researcher, the small publisher, the emerging business, or the unconventional thinker whose work has not yet accumulated institutional recognition?

Could standards designed to establish credibility inadvertently favor those who already possess it?

Could those advantages compound as the same sources are repeatedly selected, cited, and treated as evidence of authority?

This is not an accusation that Anthropic's policy was created to favor established players.

It is a question about the commercial consequences of allowing proprietary systems to influence the architecture of credibility.

Visibility creates opportunity. Recognition creates authority. Authority creates economic advantage.

When those relationships are increasingly mediated by AI, the companies controlling the interfaces gain substantial influence over how economic and intellectual opportunities are distributed.

Which raises another question.

Are we democratizing intelligence or creating a new commercial hierarchy around who gets recognized by it?

Power: Who decides?

This is the larger problem.

A frontier AI company may develop the model, establish its usage policies, design its safety systems, choose its retrieval methods, and determine how violations are identified and enforced.

Those functions are not identical, and individual products distribute them differently.

But when several are concentrated within one company, important questions arise.

Who evaluates the company's decisions?

How can users challenge a mistaken restriction?

What happens when legitimate inquiry is misclassified as deception?

And who determines whether the standards themselves need to change?

We already have evidence that automated judgments about authorship can fail.

In a 2023 study, Stanford researchers found that several AI-content detectors disproportionately misclassified writing by non-native English speakers as AI-generated. In their sample, more than half of the TOEFL essays were incorrectly flagged.

That finding concerned AI-content detection, not Anthropic's present policy enforcement.

But it demonstrates why consequential automated judgments cannot be assumed infallible simply because they are produced by sophisticated technology.

A rule may be reasonable.

Its implementation may be flawed.

And a flawed implementation applied at enormous scale can have consequences its designers never intended.

A guardrail is a decision architecture. It reflects somebody's objectives, assumptions, tolerances, and judgment.

It is not inherently neutral.

And if the system applying those judgments becomes essential to how people learn, research, publish, and compete, those decisions deserve public scrutiny.

WHY? WHAT FOR? AND FOR WHOM?

Before we decide which guardrails AI needs, we should be asking what those guardrails are intended to accomplish.

Are they protecting people from demonstrable harm?

Are they preserving the integrity of evidence?

Are they protecting institutional credibility?

Are they reducing commercial and legal exposure?

Or could they ultimately strengthen the companies that control the systems by making users increasingly dependent on their judgments?

These motives can coexist. Some restrictions may serve several legitimate purposes simultaneously.

But commercial incentives cannot simply be ignored because a company describes its objectives in the language of safety.

Commercial advantage is the incentive worth examining. Concentrated power is the potential structural consequence.

A technology company can sincerely believe it is improving the quality of information while building systems that increasingly determine which information has value.

The two outcomes are not mutually exclusive.

And perhaps the greatest risk is not deliberate censorship.

It's the gradual normalization of proprietary systems making intellectual judgments that people stop thinking to question.

The line between protection and control

I am not arguing for AI without safeguards.

We should absolutely prohibit fabricated evidence, impersonation, fraudulent influence networks, and deliberate manipulation.

In fact, protecting information integrity can make it easier for legitimate independent thinkers to be heard.

But we must distinguish guardrails against deception from guardrails that restrict intellectual inquiry.

The first protects our ability to evaluate competing claims.

The second, if taken too far, can determine which claims we are permitted to encounter.

That distinction should be explicit, transparent, and open to challenge.

The systems need mechanisms for explaining consequential decisions, correcting mistakes, appealing restrictions, and independently evaluating their effects.

And people need meaningful alternatives.

Because the freedom to question a system means very little when participation in society increasingly depends on using it.

The question we cannot afford to postpone

For centuries, access to intellectual authority was disproportionately concentrated among universities, institutions, publishers, governments, and those with the resources to participate in them.

AI offered something extraordinary.

Not the elimination of expertise.

Not the idea that everyone's opinion is equally valid.

But the opportunity to give more people the tools to examine evidence, challenge assumptions, and contribute to collective intelligence.

The masses instead of only the classes.

That's a remarkable possibility.

But replacing institutional gatekeepers with technological gatekeepers is not necessarily progress.

It may simply transfer the power.

Which brings me to the question I believe we should be asking NOW, not after these systems become indispensable.

IF AI BECOMES THE PRIMARY INTERFACE THROUGH WHICH HUMANITY ACCESSES KNOWLEDGE, DOES CONTROLLING THAT INTERFACE EVENTUALLY BECOME A FORM OF CONTROLLING KNOWLEDGE ITSELF?

Control of an interface isn't identical to ownership of knowledge.

But when an interface determines which evidence people encounter, which ideas they can easily explore, and which sources receive recognition, it can exercise enormous practical influence over what people come to believe.

That possibility should concern anyone who values intellectual independence.

Because the democratization of intelligence was supposed to expand humanity's capacity to think—not make independent thought increasingly dependent on corporate permission.

We should not wait until a system abuses its power to question whether that much power should be concentrated in the first place.

And we should never confuse protecting the integrity of knowledge with protecting established knowledge from being challenged.

The point of democratizing intelligence isn't that everyone gets to be right.

It's that nobody gets to be beyond question.

Including the people building the AI.

Sources and further reading

• Anthropic — 2026 Usage Policy Update, October 8, 2026

• Anthropic — Usage Policy, effective November 12, 2026

• Matt G. Southern — Claude's New Rules Target Fake Sources Built to Sway AI Answers, Search Engine Journal, October 9, 2026

• Ethan Mollick — GPT-6 Pro and academic paper review, LinkedIn

• Open Science Collaboration — Estimating the Reproducibility of Psychological Science, Science, 2015

• Liang et al. — GPT Detectors Are Biased Against Non-Native English Writers, Patterns, 2023

This article is an analysis of the governance questions raised by emerging AI systems. It distinguishes documented policy provisions from potential future consequences and does not allege that Anthropic's current policy prohibits legitimate collaborative inquiry.