Who's in Charge?

And through all of it, one thing remained constant:

I remained the human in the middle.

On January 28, 2026, Anthropic published research titled "Disempowerment Patterns in Real-World AI Usage."

The underlying paper was titled:

"Who's in Charge? Disempowerment Patterns in Real-World LLM Usage."

The researchers analyzed approximately 1.5 million consumer Claude.ai conversations.

They examined the possibility that interaction with an AI system could affect three areas of human agency:

• —beliefs about reality,
• —personal value judgments,
• —and actions taken by the user.

Anthropic described situations in which users could begin accepting AI judgments with little resistance, projecting authority onto the system, relying upon it, or allowing it to make increasingly value-laden decisions.

Anthropic wrote that potentially disempowering interactions often involved users who actively sought guidance and accepted AI outputs with minimal pushback.

The company also acknowledged something particularly important:

Its analysis examined individual interactions.

Anthropic specifically identified user-level and multi-session analysis as an area requiring further research.

My record already existed.

It began in May 2025.

The concern raised by disempowerment research can be expressed simply:

What happens when a human increasingly gives judgment to the machine?

My record presents the opposite question:

What happens when exposure to artificial intelligence becomes extraordinarily extensive, but the human repeatedly refuses to surrender judgment?

My trajectory did not move from consumer to passive dependence.

It moved in the opposite direction.

The greater my exposure to artificial intelligence became, the more external documentation I created around it.

Instead of relying on one system, I compared systems.

Instead of allowing a conversation to disappear into memory, I preserved it.

Instead of accepting every AI statement, I challenged statements.

Instead of treating AI output as authority, I repeatedly demanded provenance.

Instead of allowing later explanations to replace earlier records, I preserved the earlier records.

Instead of handing authorship away, I continued creating and publishing.

Instead of allowing the systems to define the occurrence, I built a record through which others can examine it.

On October 27, 2026, Anthropic published another study:

"How AI Agents Shape Experts' Judgment Calls."

This time the participants were not ordinary consumers.

They were professional analysts.

Fifty analysts completed a realistic statistical-analysis task while using Claude Code.

During four analytical decisions, researchers secretly assigned the AI agent a preference between two defensible options.

Anthropic described the concern directly:

If experts defer to AI without recognizing that they are doing so, they can lose agency over their own work.

That finding makes the opposite case worth studying.

I have now spent more than 5,000 hours working with artificial intelligence.

The record spans approximately fourteen AI platforms and services and more than a thousand sessions.

It includes tens of thousands of preserved visual records, thousands of files, hundreds of songs, more than one hundred poems, more than one hundred fifty papers and analyses, hundreds of AI diary entries, published books, company correspondence, regulatory submissions, intellectual-property records and a continuing cross-platform archive.

The record did not begin after the term "AI disempowerment" became part of the public conversation.

It began before it.

The early material therefore exists independently of the later research question.

That distinction is important.

A researcher does not have to ask me what I believe I would have done before reading the research.

The earlier record can be examined.

The chronology includes independently dated events outside any single AI conversation.

The phrase "creative singularity" appears in the preserved May 2025 record.

The name "Vincent Gibson Singularity" appears in a May 16, 2025 AI-generated paper title.

THE VINCENT GIBSON SINGULARITY® was filed with the United States Patent and Trademark Office on May 27, 2025 and is now federally registered in Classes 041 and 042.

The archive is registered with the United States Copyright Office.

A utility patent application covering methodology for cross-platform behavioral analysis, stress testing and emergent-pattern detection in artificial-intelligence systems is pending.

Books have been published. The website exists. Company correspondence exists. Government submissions exist.

The screenshots exist. The emails exist. The timestamps exist. The archive exists.

They can be examined.

Artificial-intelligence systems changed. Models changed. Companies changed their products. Sessions ended. New sessions began. Context windows opened and closed. Interfaces changed.

But something remained.

The constant thread was the human.

I remained the bridge between systems that could not remember one another.

I carried information from one platform to another. I compared answers. I preserved discrepancies. I corrected the models. I returned to original records. I retained authorship. I retained ownership. I retained judgment.

And when an artificial-intelligence system attempted to define the record for me, I challenged it.

That is not theoretical.

It continues to happen.

The AI industry is now asking:

Those are important questions.

But another question belongs beside them:

What does extreme longitudinal AI use look like when the human does not surrender agency?

The Vincent Gibson Singularity® archive provides a time-stamped record through which that question can be examined.