Editorial standards for AI assurance.

01 / Purpose & scope

What does InfoSecured publish?

InfoSecured is an independent research and practitioner resource for AI assurance, governance and risk. Its work connects claims about AI systems with the evidence, controls, oversight and decisions needed to examine them.

These standards apply to articles, explanatory guides, source directories, framework comparisons, worked examples and downloadable review tools. Each publication should answer a real reader question and make its purpose, intended use and evidence basis clear.

02 / Sources & evidence

How should sources be selected and checked?

Prefer the source closest to the claim: an issuing body’s framework, an official legal text, an original study, a technical specification or a reproducible test record. Use secondary analysis to add context, while making its role clear.

• Check the source itself. Confirm what it says, its publication or revision date, and whether it actually supports the statement being made.
• Record relevant scope. Identify versions, jurisdictions, affected systems, study conditions and other boundaries that could change the conclusion.
• Link consequential claims. Give readers direct references for important definitions, numbers, requirements and technical assertions.
• Assess evidence quality. Consider methods, limitations, conflicts, uncertainty and contradictory findings. A vendor’s statement is a claim to examine, not proof of effectiveness.
• Handle unavailable sources honestly. Do not imply that a paywalled standard or inaccessible document was reviewed in full when only a public summary was available.

Our Frameworks & Standards directory provides routes to authoritative references. The issuing source remains the place to confirm its current status and wording.

03 / Claims & interpretation

What does the evidence support?

A conclusion should be no broader than the evidence behind it. Identify what was examined, under which conditions, and what remains uncertain. Keep these three forms of content distinct:

Source-backed fact

A statement supported by a cited document, observation or recorded result. Preserve the source’s relevant qualifications.

InfoSecured analysis

An interpretation, recommendation or mapping that explains its reasoning. A practical crosswalk is not an official declaration that two frameworks are equivalent.

Illustrative example

A scenario, sample record or synthetic dataset used to explain a method. Label it where it appears; do not present it as a client case or measured finding.

Claims of compliance, certification, validation, regulatory approval or successful deployment require specific supporting evidence and a defined scope. Completing a checklist or organizing an evidence file does not establish those outcomes.

04 / Original work & attribution

What makes a publication worth reading?

Original value can come from a documented experiment, an evidence-review method, a carefully reasoned comparison, a useful decision tool or a worked application of existing guidance. Explain the contribution instead of repackaging source material with different wording.

Credit the people and publications whose work informs the analysis. Mark direct quotations, preserve their meaning and keep them proportionate. Cite borrowed methods, data and substantive ideas; respect the permissions and licenses governing reused material.

When reporting original tests or calculations, describe the method, inputs, conditions and limitations sufficiently for readers to assess the result. Share supporting material where it is lawful and appropriate, without exposing confidential information.

Programmatic publishing should produce distinct answers based on distinct evidence or analysis. Do not create near-identical pages for keyword variations or substitute volume for a useful contribution.

05 / AI & automation

How may AI assist the editorial process?

AI may assist research discovery, organization, drafting, editing, formatting and publication preparation. Responsibility for the finished work remains with InfoSecured and the people reviewing it.

• Verify against evidence. Check generated claims, quotations, references, calculations and summaries against the actual source material.
• Retain human review. An automated critic can flag problems; it does not replace accountable editorial judgment before publication.
• Review the whole publication. Apply the same accuracy standard to titles, descriptions, image captions, alternative text and structured data.
• Protect sensitive inputs. Do not submit confidential, personal or restricted material to an external tool without appropriate authorization and handling safeguards.
• Explain consequential AI use. Disclose AI-generated evidence, synthetic examples or a material role in the method wherever that information affects how readers should interpret the work.

Automated publishing and updates should follow an approved content version. Changed factual claims or conclusions require renewed review; a generated citation or another model’s agreement is not verification.

06 / Publication review

What should be checked before publication?

Review should connect the reader’s question, the underlying evidence and the final explanation. The depth of review should reflect the consequences of getting the subject wrong.

1. Define the question and contribution.

   Identify the reader, purpose and scope. Confirm that the publication adds a useful answer, method, example or analysis.

2. Check the factual basis.

   Verify important claims, source versions, names, quotations and results. Examine gaps and contradictory evidence.

3. Edit for clarity and judgment.

   Separate facts from interpretation, state relevant limits, remove unsupported authority claims and make the practical implications readable.

4. Check the delivered page or file.

   Test links, headings, accessibility, metadata and downloadable resources. Confirm that examples are labeled and the published version matches the reviewed content.

5. Identify accountability and review triggers.

   Record the publication version and responsible reviewer. Revisit material when evidence, source requirements, technology or a substantive reader correction changes its basis.

A named author, contributor or reviewer should have the role described. Do not imply external peer review, independent assurance or expert endorsement unless that work actually occurred.

07 / Practical resources

How should review tools and examples be used?

InfoSecured resources help structure questions, evidence and decisions. The AI Assurance Evidence Review Kit, for example, provides a connected workbook for organizing an assurance review.

A blank template establishes a structure. Its completed records become useful evidence when they are linked to real systems, relevant tests, accountable owners and documented decisions. Review the substance of those records as well as whether the fields are complete.

• Choose the resource that fits the system, review question and applicable requirements.
• Check the version, instructions, assumptions and worked examples before using it.
• Keep illustrative data separate from actual evidence; protect sensitive records and apply suitable access and retention controls.
• Use qualified stakeholders to assess interpretations or decisions with legal, regulatory, audit or operational consequences.

Resources support preparation and review. They do not, by themselves, provide a legal opinion, audit opinion, certification or an organization-specific compliance determination.

08 / Independence & disclosures

What relationships should readers know about?

Disclose sponsorship, paid placements, affiliate links, vendor-provided access and other interests that could reasonably affect a reader’s assessment of a publication. Place the disclosure with the relevant content or link, where it can inform the reader’s judgment.

Commercial involvement must not justify unsupported findings or concealed limitations. Comparisons should explain their criteria and avoid implying comprehensive testing when only a limited selection was examined.

Case material, personal information and confidential evidence require appropriate permission and protection. Identify anonymized or altered examples accurately, and do not imply that a composite scenario describes a verified real-world engagement.

09 / Corrections & updates

How can readers report an error?

Contact InfoSecured with the page URL or resource name, the passage or field in question, and the source or explanation supporting your correction. Please avoid sending sensitive records through a general inquiry.

A reported issue should be checked against the underlying evidence. The appropriate response may be a factual correction, clearer scope, a revised example or file, an outdated-content notice, or withdrawal of an unsupported claim.

• Material corrections: identify the changed finding, conclusion or instruction and explain the correction near the affected content or in a revision note.
• Resource revisions: distinguish the new version and describe changes affecting use or interpretation. Make clear which version readers should use.
• Routine edits: improve spelling, formatting or readability without implying that the underlying evidence was reassessed.
• Review dates: change a source-review date only after rechecking the relevant sources. Record substantive revisions without making older content appear newly researched.

A correction should make the record more reliable and help readers understand whether earlier conclusions or decisions need reconsideration.

10 / Common questions

Quick answers about these standards.