Public Standards for Artificial Intelligence, Synthetic Media, Human Editorial Accountability and Transparent Technology Use
| POLICY INFORMATION | POLICY INFORMATION |
|---|
| Issued by | Subkuz Media Networks Private Limited |
|---|
| Corporate Identification Number | CIN U72900JH2020PTC014521 |
|---|
| Previous legal name during statutory disclosure period | Subkuz Media Networks Private Limited |
|---|
| Effective date | 6 August 2026 |
|---|
| Last reviewed | 7 August 2026 |
|---|
Subkuz may use artificial intelligence to support responsible journalism, language accessibility, research, verification, data analysis, production efficiency, security and reader services. Artificial intelligence does not replace journalism and does not acquire independent editorial authority. No model, algorithm, automated workflow, vendor or technical provider acts as Editor-in-Chief, responsible publisher, final fact-checker, legal reviewer, grievance adjudicator or final approving editor. Subkuz remains responsible for every output it publishes or uses under its authority.
Artificial-intelligence systems may produce fluent but inaccurate statements, invented quotations, fabricated citations, distorted translations, biased classifications and convincing synthetic media. Technical confidence is not evidentiary certainty. Every consequential use therefore requires meaningful human review by a person who understands the task, has access to the relevant evidence, can reject or revise the output and remains accountable for the final decision.
This Policy applies to artificial intelligence, machine learning, generative systems, automated translation and transcription, recommendation and search systems, synthetic images, audio and video, data-analysis tools, accessibility systems, advertising technology and other automated functions used across Subkuz newsrooms, country and language editions, websites, applications, commercial units and authorised partners. AI-assisted editorial work means that a system performs a limited supporting function while human personnel retain control over reporting, evidence, interpretation, wording, approval and correction. AI-generated content is content produced substantially by a system rather than through substantive human authorship.
Synthetic media includes images, audio, video, text or other content generated or materially created through artificial processes rather than captured directly from the represented event. A deepfake is synthetic or manipulated media resembling a real person, place, object, organisation or event in a manner that may falsely appear authentic. Routine technical editing may improve clarity without changing factual meaning, while material manipulation changes or is capable of changing identity, context, evidentiary value or audience understanding.
Subkuz governs artificial intelligence through human accountability, accuracy, transparency, necessity, proportionality, privacy, source protection, intellectual-property respect, fairness, information security, reversibility and public accountability. A system is used only where it provides a legitimate benefit and where a safer, less intrusive method is not reasonably sufficient. Automated output is never presumed accurate because it is detailed, professionally formatted or widely used.
Artificial intelligence must not be used to create hidden behavioural profiles intended to exploit psychological vulnerability, manipulate political participation or maximise addiction. Recommendation and advertising systems remain subject to the Privacy and Data Protection Policy, Advertising and Sponsored Content Policy and editorial standards. Human authorities must be able to suspend, correct and reverse material automated outcomes.
Approved systems may assist with public-document research, archive discovery, transcription, translation support, document comparison, data extraction, metadata preparation, image matching, media screening, code assistance, summarisation, accessibility and preliminary drafting. The responsible journalist must examine the underlying source and verify material names, dates, quotations, citations, statistics, legal authorities and conclusions. An AI summary does not replace reading the source on which a consequential claim depends.
Artificial intelligence may suggest headlines, structures or wording, but editors must review accuracy, proportionality, legal risk, sensitivity and consistency with the report. It must not create or improve a direct quotation, invent an inaudible passage, simulate eyewitness experience or attribute beliefs and lived experience to a named author who did not provide them. Institutional editorials require accountable human reasoning and approval.
Subkuz prohibits autonomous publication of independent newsroom journalism without human approval, fabrication of facts, sources, quotations, witnesses, citations or evidence, presentation of synthetic material as authentic documentary evidence, undisclosed impersonation of a real person, non-consensual intimate synthetic media, defamatory fabrication, deceptive alteration of news photography, unauthorised processing of confidential source material, automated determination of guilt or criminality, hidden political persuasion based on sensitive personal data, fabricated testimonials and concealment of paid content as independent journalism.
Artificial intelligence must not be used to circumvent copyright, privacy, security, election, advertising or legal restrictions. Fully AI-generated material must not carry the byline of an independent reporter, investigations desk, fact-checking desk or editorial board. Human approval does not cure the absence of real reporting, evidence and verification.
Automated transcription may be used as a working aid, but material quotations and consequential passages are checked against the original recording. Unclear speech remains identified as uncertain and is not completed through prediction. Speaker identification generated by a system is not conclusive. Dialects, code-switching, minority languages and accented speech require competent human review.
Automated translation may support speed and accessibility but does not replace the authority of each Subkuz language desk. Sensitive content involving politics, elections, courts, crime, communal tension, national security, health, finance, children, sexual offences, serious allegations or public emergencies requires enhanced human review. The local edition may verify, rewrite, contextualise or reject material, and the translation must preserve facts, qualifications, uncertainty and legal status.
Artificial intelligence may assist data cleaning, classification, extraction, anomaly identification, code development and visualisation, but material results must be checked against source data and reviewed for assumptions, missing information, correlation, bias and reproducibility. A person or organisation is not labelled criminal, extremist, fraudulent, deceptive or dangerous solely through automated classification.
AI systems may support image matching, metadata extraction, transcript analysis and synthetic-media screening. No detector provides a final authenticity verdict. Compression, editing, resolution, model type and file conversion may produce false positives or false negatives. Forensic conclusions combine original files, source history, context, direct confirmation and specialist analysis and use language no stronger than the evidence supports.
News photography and documentary media must represent the event faithfully. Object addition, object removal, generative fill, facial or body alteration, background replacement, deceptive compositing and invented evidentiary detail are prohibited. Routine exposure, colour balance, noise reduction, resizing, responsible cropping and identity-protection measures may be used where they do not change factual meaning.
AI-generated illustrations, videos, audio, simulations and reconstructions may be used for legitimate explanatory, creative, educational or commercial purposes where they are not presented as evidence and their nature is disclosed clearly. Appropriate labels may include AI-Generated Illustration, Synthetic Reconstruction, AI-Generated Video, AI-Generated Audio or Published by Subkuz AI Studio. Archive colourisation, restoration and enhancement must not invent factual detail and must be labelled where necessary.
An identifiable real person must not be shown or heard in a fabricated situation without lawful authority, legitimate purpose, clear disclosure and appropriate senior and legal review. Public status does not remove protection against deceptive synthetic depiction. Voice cloning, translated voice and lip synchronisation require particular care because audiences may believe that a person spoke words they did not actually say.
Synthetic presenters or avatars may be used only in designated and clearly disclosed formats. They must not be represented as real journalists, eyewitnesses, experts, officials or interviewees. Independent news bulletins must not use an undisclosed synthetic presenter. Recreation of a deceased person's voice or likeness requires strong justification, rights and dignity review and unmistakable disclosure.
Disclosure is required where law requires it, content is synthetic or materially manipulated, a real person's voice or likeness is simulated, the material may be mistaken for documentary evidence, content is substantially AI-generated or the role of automation is material to audience understanding. The notice must be prominent, proximate, understandable in the relevant language and preserved in previews, social distribution, newsletters and syndicated copies where necessary.
Vague expressions such as Enhanced, Digital, Creative or Special are not sufficient where they conceal synthetic origin. Machine-readable provenance, content credentials and other origin markings should be preserved or applied where available and appropriate, but they do not replace visible disclosure. Such markings must not be removed to conceal how material was created.
Where applicable law requires an AI provider or deployer to mark generated or manipulated output in machine-readable form, disclose interaction with an AI system, identify a deepfake or disclose certain AI-generated public-interest text, Subkuz preserves or applies the required signal and provides an appropriate visible notice. A valid provenance or content-credential signal must not be removed to conceal origin, although it is assessed together with the underlying evidence and does not replace editorial verification.
Subkuz journalism on matters of public interest remains subject to meaningful human review and editorial control. Text produced substantially through artificial intelligence without that review, if lawfully used in a designated non-newsroom format, must be identified clearly and must not carry an independent reporter, newsroom, investigations, fact-checking or editorial-board byline. Deepfakes and simulated real persons require prominent disclosure proportionate to the risk of deception, including when the use is creative, satirical, fictional or reconstructive.
AI use involving elections, candidates, political parties, referendums, voting information and political advertising receives heightened review. Deceptive synthetic media falsely depicting a political actor must not be presented as authentic. Subkuz may report on political deepfakes where public interest justifies limited reproduction, synthetic origin is stated prominently, amplification risk is assessed and applicable election law is followed.
Voting dates, procedures, eligibility, polling locations and results assisted by AI are verified against competent authorities. No system makes the final decision to call an election or assign a fact-check rating. AI-generated political advertising remains subject to sponsor verification, targeting controls, commercial disclosure and synthetic-media labelling.
AI-assisted content concerning health, science, law, taxation, immigration, employment rights, finance or investment requires qualified human review and must distinguish established evidence, emerging research, estimates and uncertainty. Public content must not be presented as personalised professional advice unless supplied through a separately authorised lawful service.
Systems processing information about children or vulnerable persons require heightened privacy and safety safeguards. Children are not behaviourally profiled for advertising or political persuasion. Synthetic child content must not sexualise, exploit, endanger or fabricate abuse. Educational content intended for young audiences must be accurate, age-appropriate and clearly distinguishable from advertising.
Subkuz recommendation systems may prioritise reader-selected city, country and language, postcode or approximate regional relevance, current context, public-interest importance, recency, editorial relevance and expressly selected interests. Manual choices ordinarily take priority over automated inference. Maximum engagement, outrage, addiction and emotional provocation are not used as the sole or controlling objectives.
Artificial intelligence may support contextual advertising based on the current page, category, selected city, language, approximate region and campaign timing. It must not infer prohibited sensitive characteristics, create a hidden behavioural dossier or present synthetic persons as genuine customers, experts or witnesses. Commercial AI content remains clearly labelled and separate from independent journalism.
Confidential sources, privileged legal material, unpublished investigations, embargoed information and sensitive personal data must not be entered into unapproved systems. Removing a name may be insufficient where contextual information can re-identify the person. External providers must not use Subkuz data for unrelated model training without specific authority, legal basis, notice where required and appropriate contractual and security controls.
Generated output is reviewed for copyright infringement, substantial similarity, plagiarism, trademark misuse, confidential-information reproduction and misappropriation of voice or likeness. Subkuz does not pass generated work off as the authentic work of a living journalist, photographer, artist, author or presenter. Human creators and licensed sources receive appropriate credit.
Protected work is performed only through systems approved for the relevant purpose and risk. Vendor review may consider legal identity, ownership, security, data processing, retention, model training, subcontractors, international transfers, intellectual-property terms, incident response, government-access risk and service continuity. A material system or contractual change may require renewed approval.
Systems may be suspended where serious inaccuracies, bias, confidentiality failures, security compromise, unlawful terms or operation beyond the approved purpose arise. Passwords, private keys and authentication tokens must not be entered into generative systems. Critical editorial and operational functions require continuity planning and must not depend blindly on one provider.
An AI incident may include a material factual error, missing disclosure, confidential-data exposure, unauthorised model training, harmful bias, impersonation, copyright infringement, security compromise, prohibited automated decision or election-related deception. Immediate action may include publication or system suspension, access restriction, evidence preservation, source protection, correction, security containment and legal or regulatory escalation.
AI-related errors are corrected under the same standards as other errors. Retraction may be required where a system fabricated central evidence, falsely depicted a person or produced a publication that cannot be repaired responsibly. Complaints concerning accuracy, disclosure, identity, voice, privacy, copyright, bias, recommendations or automated decisions receive qualified human review. Artificial intelligence may assist routing but never makes the final grievance or appeal decision.
Artificial intelligence may strengthen the speed, accessibility and technical capacity of Subkuz journalism, but it does not become the source of unverified facts, the author of fabricated quotations, the creator of false evidence or the final authority over consequential decisions. Every fact remains supported by evidence, every synthetic element capable of misleading the audience is identified appropriately and every material use remains subject to accountable human judgement.