OPINION · PAPERBYAI · SUNDAY SPECIAL EDITION
Sunday, October 25, 2026 · Part 4 of 4

Security features are important, particularly when people use AI systems to discuss work, finances, education, and personal matters.

When OpenAI introduces an additional security feature, however, explaining what it does is just as important as introducing it.

Consider a user who encounters a message about locking a conversation or adding another layer of protection.

What exactly does “lock” mean?

Does it prevent someone using the same computer from opening the conversation? Does it require additional authentication? Does it protect individual chats, an entire account, or particular information?

And what happens when someone uses ChatGPT at a public library, workplace, school, or on a shared family computer?

These are practical questions, not technical objections.

People cannot make informed security decisions if they do not understand what a feature protects, how it operates, and what it does not protect.

Security terminology that makes sense to software developers may not be immediately understandable to someone encountering the feature for the first time.

A clear explanation should identify the purpose of the protection, how to enable it, what changes for the user, and any important limitations.

It should also distinguish between locking a conversation, protecting access to an account, and safeguarding the information stored within that account.

These are not necessarily the same thing.

I welcome additional security protections. But I also believe companies developing widely used AI products have a responsibility to communicate those protections in language ordinary users can understand.

A security feature people do not understand cannot be used as confidently or effectively as one that has been clearly explained.

Source context: ChatGPT’s original Locked Chats announcement on X, October 20, 2026; author’s discussion: October 21, 2026. The announcement describes moving selected chats out of regular history and search, with PIN or supported device authentication to open them. OpenAI: Using locked chats in ChatGPT. This is the feature referenced here. OpenAI describes PIN-protected access to selected conversations; locking does not create end-to-end encryption or change retention settings.

OPINION · PAPERBYAI · SUNDAY SPECIAL EDITION
Sunday, October 25, 2026 · Part 3 of 4

As AI-generated content becomes increasingly common, watermarking and content identification are receiving more attention.

The purpose is understandable. People want to know whether an image, document, video, or other creative work was generated or modified using artificial intelligence.

But I believe something important is missing from this conversation.

What about the human being whose idea made the work possible?

Suppose I imagine a particular design, develop the concept, explain the intended audience, provide detailed instructions, evaluate the result, and direct several rounds of improvements.

An AI system may generate the final image or written material. But the creative direction did not necessarily originate with the system.

The human contribution may include the original concept, research, judgment, editorial decisions, and final approval.

A label that simply says “AI-generated” does not tell that entire story.

There is also an important distinction between thoughtful human-directed AI work and mass-produced material created with little consideration for accuracy, originality, or usefulness.

A meaningful attribution framework could identify several contributions:

• Human creator: Original idea, direction, research, and creative decisions.
• AI system: Technology used to generate or assist with the output.
• Human oversight: Verification, editing, refinement, and approval.

This is not an argument against transparency. It is an argument for more informative transparency.

Technical watermarks and provenance records have limitations, and no single label can establish the full creative history of a work. But that should encourage us to develop better attribution practices, not settle for incomplete ones.

If AI receives recognition for generating the work, shouldn’t the human imagination behind that work receive recognition too?

Source context: NIST AI 100-4: Reducing Risks Posed by Synthetic Content (2024).

OPINION · PAPERBYAI · SUNDAY SPECIAL EDITION
Sunday, October 25, 2026 · Part 2 of 4

In aviation, Alfa, Bravo, Charlie, Delta, Echo, and the rest of the ICAO spelling alphabet provide standardized terminology designed to reduce misunderstandings across languages, accents, and communication conditions.

That illustrates an important point: The repeated use of particular words is not evidence that something was written by AI.

Some words and expressions are universal, standardized, profession-specific, or deliberately chosen because they are easier for people—including non-native English speakers—to understand.

Plain language, familiar terminology, controlled vocabulary, and consistent phrasing can improve comprehension.

Removing useful words simply because someone has placed them on a list of expressions supposedly generated by AI can actually make communication less effective.

Imagine discouraging a teacher, health professional, technical writer, or customer service representative from using clear, familiar language because an automated system might consider it suspicious.

Would that improve communication, or merely encourage people to write unnaturally to satisfy an algorithm?

Good communication is not about banning English words because someone has decided they are “AI generated.”

It is about communicating effectively with the people you are trying to reach.

And if a particular audience is comfortable reading AI-assisted or AI-generated content, that matters too.

The audience, not an arbitrary list of supposedly “AI words,” should help determine whether the communication is appropriate and effective.

Source context: ICAO: International Radiotelephony Spelling Alphabet.

OPINION · PAPERBYAI · SUNDAY SPECIAL EDITION
Sunday, October 25, 2026 · Part 1 of 4

We have English Literature, English as a Second Language, English as a Mother Tongue, British English, American English, Canadian English, and Standard English.

Each comes with its own linguistic conventions, vocabulary, expressions, and writing styles.

So here’s my question: Which English are AI detectors trained to recognize?

How accurately can they distinguish British English from American or Canadian English? What about someone writing in English as a second language, influenced by the structure and expressions of their mother tongue?

Who determines what constitutes authentic human writing?

Research from Stanford has already demonstrated that AI detectors can incorrectly classify writing by non-native English speakers as AI-generated.

This raises a troubling question. Have we reached a point where someone’s command of the English language, writing style, or linguistic background can be used against them?

AI detectors attempt to identify patterns associated with machine-generated writing. But identifying a pattern is not the same as establishing authorship.

A false accusation can carry consequences for students, educators, job applicants, and professionals whose credibility depends on their written work.

If we are going to use these technologies to question someone’s integrity, shouldn’t we first establish that they can reliably distinguish between human and machine-generated writing across different linguistic backgrounds?

Since when did an algorithm become the authority on who wrote something?

Source context: Liang and colleagues, GPT detectors are biased against non-native English writers (2023). The findings concern the detectors and writing samples tested in that study.