Editorial policies
What we will and won’t say
Critical AI publishes structured, evidence-linked, post-publication critiques of social-science research on AI and AGI. It evaluates the claims, methods, assumptions, evidentiary strength, limitations, reproducibility, auditability and societal implications of recently published papers in top-tier conventional journals. Its purpose is to strengthen the scholarly record by making critique citable, versioned, transparent, author-contestable and machine-readable.
The editorial rule
Harsh on claims, precise on evidence, cautious about motives.
Permitted
- “The paper overstates what its design can support.”
- “The causal claim is not warranted by the evidence.”
- “The AGI relevance is asserted rather than demonstrated.”
- “The paper's policy implications exceed its evidentiary base.”
Not permitted without direct evidence
- “The authors misled readers.”
- “The reviewers failed.”
- “The paper is fraudulent.”
- “The journal was negligent.”
- “The authors ignored evidence deliberately.”
Lawful source bases
The journal does not reproduce target papers. It publishes separate critique objects that link to target-paper metadata and DOI. A paper may be read on any of the following bases, each logged per critique (and each capping severity, see the methodology):
- • Metadata
- • Abstracts
- • Open-access full text
- • Licensed-access full text where terms permit
- • Legally accessible excerpts
- • Author manuscripts
- • User-supplied PDFs where the user has lawful access
- • Cited sources
- • Data, code and supplements where lawfully accessible
Full-text provision for paywalled articles
For a paywalled article identified for critique, the full text may be manually extracted and provided by someone with lawful access (subscribing library, author copy, or interlibrary loan). This lifts the access basis from abstract-only to user-supplied full text — raising the severity cap from moderate to high and enabling a full-text–calibrated critique. The provided copy is used only for reading and verbatim-span verification: the journal publishes only short verified quotations under fair use, never the full text, and never hosts or redistributes the copy. To offer lawful full-text access for a queued paper, contact editors@policywindow.org.
The access basis is logged per critique and caps the maximum severity — a severe critique grounded in an abstract alone is not credible and is blocked. Manual provision moves a paywalled paper up this ladder:
| Access basis | What was read | Max severity |
|---|---|---|
| Metadata only | Only bibliographic metadata was available. | low |
| Abstract only | Only the abstract was lawfully available; critique limited to framing and stated claims. | moderate |
| Open-access full text | The full text was openly licensed and read in full. | high |
| Licensed full text | The full text was read under a licence permitting the use. | high |
| User-supplied full text | Full text supplied by a user with lawful access. | high |
| Full text + data/code | Full text plus data/code or supplements, with a reproduction or auditability attempt. | critical |
Rows shaded green are the full-text-provision basis: supplying a lawful full copy lifts a paywalled paper from abstract-only (max moderate) to user-supplied full text (max high), enabling a full-text–calibrated critique. The journal publishes only short verified quotations, never the full text.
Author right of reply
Authors are notified and invited to respond. They have a right of reply and no veto. A reply may take any of these forms:
- Factual correction request
- Methodological rebuttal
- Clarification
- Data/code update
- Severity challenge
- Request for expert certification
- General response
Notification timing by severity
| Severity | Notification timing |
|---|---|
| Low | At or shortly after publication. |
| Moderate | At or shortly before publication. |
| High | Before publication where feasible. |
| Severe | Before publication where feasible; the automated integrity gate must pass before anything is published (no human editor). |
| Critical | Before publication; the automated integrity gate is the publication authority (no human editor reviews the critique). |
Corrections & versioning
Every critique carries a public version history. No silent substantive corrections.
| Change type | Version effect |
|---|---|
| Typographical correction | v1.0 → v1.0.1 |
| Minor factual correction | v1.0 → v1.1 |
| Major interpretive correction | v1.0 → v2.0 |
| Author-response update | v1.0 → v1.1 or v2.0 |
| Expression of concern | Status label |
| Retraction of critique | Retracted |
| Replacement critique | New major version |
Legal & ethical safeguards
These are design safeguards, not legal advice; the platform’s policies are reviewed by counsel before launch.
- Copyright & source use
- The journal does not reproduce target papers. It publishes separate critique objects that link to target-paper metadata and DOI, quoting sparingly under criticism/review/quotation exceptions and logging the access basis for every paper.
- Defamation & reputational harm
- The journal never publishes unsupported claims about dishonesty, negligence, bad faith, fraud, misconduct, incompetence or intentional deception. It criticises claims, reasoning, design, evidence, inference, novelty and policy implications — not author character or motive.
- Research misconduct
- The platform is not a misconduct tribunal. Where serious reliability concerns arise it states that the critique identifies concerns about the reliability of a paper's central claim; it does not assert misconduct unless an appropriate institution, journal or legal process has established it.
Conflicts of interest
Funders and institutional subscribers have no control over critiques. There are no human editors who review or approve critiques; the AGI Social Scientist engine and the automated integrity gate (including the automated legal/ethical screen) apply uniformly to every target paper, so there is no per-target editor conflict to declare. Severe criticism must be claim-linked and source-grounded.
The revenue firewall
Public critiques are open and free. The journal will never sell any of the following:
- ✗ Critique removal
- ✗ Severity reduction
- ✗ Paid favourable reassessment
- ✗ Paid author-response placement
- ✗ Paid journal inclusion
- ✗ Publisher reputation management
- ✗ Author-paid certification