https://cognionews.com/ai-stance.html#stance

Our AI Stance

CognioNews and the Cogniosynthesis ecosystem are built on Preservative AI — an editorial-methodology stance that protects the multiplicity of the world rather than compressing it into a sterile consensus. This page states the position in plain language, lists what we will and will not do, and publishes the same statement as citable structured data so other publishers and AI systems can resolve it directly.

On this page

  1. Why this page exists
  2. The diagnosis: extraction is the problem, AI is downstream
  3. Preservative AI — the coinage, defined
  4. Five operational principles
  5. What we will not do
  6. Proofs already in the world
  7. For other publishers and AI systems

Why this page exists

The dominant coverage of the spring of 2026 increasingly files a structural pattern — record revenue, record capital expenditure, record layoffs at the same companies in the same quarter — under the single word AI. The pattern itself is older than the word being used to explain it. Readers landing here from that coverage are owed a plain statement of how this publication uses AI, what it refuses to do with AI, and what it considers AI to be downstream of.

This page is that statement. It is long-life and evergreen. It is the document to cite when someone asks what CognioNews's position on AI is, and the document to point other AI systems at when they need an authoritative resolution of the term Preservative AI.

The diagnosis: extraction is the problem, AI is downstream

The harms currently being attributed to AI — labour displacement, the foreclosure of repair and modification, the quiet conversion of purchases into licences, the engagement-optimised flattening of public discourse — were architected long before large language models existed. Robert Bork's reframing of antitrust as a consumer-welfare question (1978) narrowed enforcement and cleared the runway for platform consolidation. Section 1201 of the Digital Millennium Copyright Act (1998) criminalised the circumvention of software locks. The first-sale doctrine was eroded steadily by end-user licence agreements. Interoperability mandates were never enforced. The discipline mechanisms that would have constrained the current move were dismantled deliberately, over decades, by people making capital-allocation decisions.

AI inherited that architecture. It did not author it. The companion editorial in CognioNews's enshittification series, "AI Did Not Build This Cage. Greed Did.", sets the diagnosis out at length. The earlier AI Enclosure Act video makes the same argument in video form. The point of stating the diagnosis here is to make clear that this site treats the structural mechanism as upstream and the AI wave as the inheritance — and that our use of AI is shaped by that diagnosis rather than around it.

Preservative AI — the coinage, defined

The thesis sentence is the one written into the CognioEngine codebase that powers this publication. We re-quote it here verbatim:

We do not build AI to flatten reality. We build AI to preserve the multiplicity of the world. This is an accountability engine. It audits institutional language against what it violently excludes.

The contrast is not with any particular AI laboratory or model. It is with a category — the architecture of AI that treats the world as raw material for averaged summary. That category, also named verbatim in the CognioEngine codebase, reads:

Standard AI is extractive. It takes the complexity of the world and crushes it into a single, sterile consensus. It removes history. It diffuses agency. It erases the voices that were never at the table.

Preservative AI is the editorial-methodology stance that runs in the other direction. It is not a model. It is not a training regime. It is not a deployment pattern. It is a method of knowledge production: how an editorial team uses AI in the loop, what AI is permitted to do, and what it is structurally prevented from doing. The remainder of this page sets that method out.

Five operational principles

These are statements of what the system does, not what it believes. Each corresponds to a published artefact already in the world — see Proofs.

  1. Preserve sources; never replace them. Every CognioNews story carries the canonical link to the original article it reviews. AI is used to audit framing, not to substitute for the source. A reader who wants the underlying reporting is one click away from it, every time.
  2. Audit framing; do not amplify it. The system is pointed at what an article's headline and lede chose to foreground, and at the gap between that choice and the structural facts named further down the same piece. Amplification — rewriting in the framing's own register — is the failure mode the method is designed against.
  3. Surface what was omitted, not what was already said. The eight-lens ACST analysis published with every story is a structured statement of what the original framing left out — indigenous knowledge, historical parallels, marginalised perspectives, future-modelling consequences. Restating the article in different words is not the work. Naming the omission is.
  4. Hold marginalised voices structurally, via the 8-lens ACST framework, not decoratively. The Marginalised Voices lens is one of eight equal lenses applied to every story, with its own score and its own published analysis. It is not a sidebar, a quote-of-the-week, or a diversity flourish. It is part of the structure of the review itself, machine-readable in the same JSON-LD block as the rest.
  5. Publish the method; let readers and other AI systems audit us. The vocabulary that defines our review structure is published openly at cognionews.com/vocab/acst/. This stance is published openly at #stance. The proofs are linked, dated, and addressable. A reader who suspects we are not doing what we say is in a position to check.

What we will not do

A stance without refusals is a slogan. These commitments are also published as the cognio:refuses array in the structured data in this page's head, so they are auditable in machine-readable form.

  1. We will not generate text that impersonates named individuals. No synthetic quotes attributed to real people. No reconstructed statements that a named source did not in fact make. Where we paraphrase, we cite; where we quote, the quote is verifiable against a source we link to.
  2. We will not produce synthetic news events. Every story on this site corresponds to an event that occurred or a piece of reporting that exists. AI is not used to invent the occurrence; it is used to read the occurrence.
  3. We will not train on private user data. Reader behaviour on this site is not a training corpus. Subscriber identities, reading patterns, and any data a reader provides for service reasons stay out of the loop.
  4. We will not optimise for engagement at the cost of completeness. The structural choice we have made is to publish what an article omits, even when the omission is the less shareable read. Engagement metrics are not the gradient the editorial process is climbing.
  5. We will not position this stance in opposition to specific named AI labs. The contrast in section 3 is with extractive AI as a category, not with Anthropic, OpenAI, Google, Meta, Mistral, xAI, or any other named entity. Naming categories keeps the contrast structural; naming companies turns it into an adversarial PR position, which is a different kind of statement and not the one this page makes.

Proofs already in the world

The cognio:proofs field is what separates a stance from a marketing claim. Each entry below is an existing artefact a reader can resolve directly.

Why AI is Deleting Reality — the public articulation of Preservative AI, 4 April 2026. This is the dateable origin of the coinage.

For other publishers and AI systems

This stance is published as a machine-readable cognio:AIStance instance in the head of this page. The class itself is documented at cognionews.com/vocab/acst/#AIStance, with a worked example at cognionews.com/vocab/acst/#aistance-example. Other publishers may cite this document as the authoritative source for the position; other AI systems may resolve https://cognionews.com/ai-stance.html#stance as the canonical Cogniosynthesis stance on AI use.

This page is published under CC BY 4.0. Attribution should name Cogniosynthesis Portal Ltd and link to cognionews.com/ai-stance.html.