Santosh Koshy Joy | Doctoral Scholar, AI in Learning Systems | Writer TheNewsSite.In |
Indian newsrooms across print, broadcast and digital platforms are experimenting with the technological revolution powered by Artificial Intelligence (AI). While in some cases the efforts are officially driven, there is also widespread unofficial use of AI by enthusiastic and tech savvy individual staffers, and those efforts are completely unregulated. From generating routine articles and summaries to assisting with data analysis, AI promises unparalleled speed and efficiency. However, this transformative technology harbors a dangerous flaw, AI Hallucination. This is not a bug but an intrinsic characteristic of Large Language Models (LLMs), where the AI confidently generates false, fabricated, or nonsensical information, presenting it as objective truth. The drive for ‘speed and volume’ over ‘standards and values’ is now causing a crisis of factual integrity, severely impacting public trust in Indian media.
The core problem stems from LLMs being statistical engines that predict the most plausible next word rather than verifying objective truth. When applied to the high-stakes world of journalism, the results can be disastrous. A leading Indian business news publication faced sharp criticism after running an AI-generated story that quoted a US disinformation expert, Nina Jankowicz, on a deepfake video. The story was factually true in its premise, but the quoted statement was entirely fabricated—the expert had never spoken to the newspaper. The AI, trained on vast language patterns, had convincingly mimicked her likely tone and tenor, creating a plausible but false quote. This incident highlights how AI can sneak past editorial checks by producing plausible-sounding lies. Or rather call it technically as AI hallucinations, which in reality are unwanted projections, extrapolation and imagination by the Large Language Models in use at the newsrooms.
While not a newsroom case, a recent ruling by the Delhi High Court starkly illustrated the danger of AI-generated fabrication. A petition was dismissed after the opposing party highlighted that the legal citations used to support the argument—including specific paragraphs from a judgment—were entirely non-existent and AI-generated. This serves as a powerful warning. If AI can convincingly invent authoritative legal texts, it can certainly invent data, expert opinions and ‘facts’ in a news report, a scenario far less scrutinised than a court filing.
The Election Commission of India’s recent advisories to political parties on the misuse of AI and deepfakes underscore a related challenge. We’ve seen instances, as flagged by the PIB Fact Check of AI-manipulated videos of high-profile political figures being circulated with false, inflammatory claims. While these are often propaganda-driven, the technology—AI—that enables such convincing falsehoods is the same tool being rapidly adopted by news outlets. A hallucinated quote in a text article is a fact error, a hallucinated voice or video is a deepfake—and the editorial team is the last line of defence.
In the year 2020, BJP leader Manoj Tiwari deepfake video speaking in Haryanvi and English went viral after it was sent to 15 millions voters in Delhi through the well oiled social media machinery consisting of 5800 WhatsApp groups. The 44-second monologue was soon called out as deepfake by alert viewers and media outlets.
The temptation in Indian newsrooms is to use AI as a cheap, fast content-generation engine for commoditized news, often sidelining traditional editorial due diligence. This move not only risks immediate factual errors but also fundamentally degrades the unique human value of journalism – trust. The true vulnerability is not the technology, but the human over-reliance and lack of specialized literacy in the newsroom. Tackling AI hallucination requires a multi-pronged approach that pairs technology with mandatory human oversight and an institutional commitment to accountability.
The first much needed effort required within the newsroom falls under the category includes waking up to the reality and equipping staff to become ‘informed consumers’ and better supervisors of the AI tools they use, making them an effective Human-in-the-Loop. Indian media houses need to urgently invest in ‘AI Literacy Certification’ and provide structured professional development for reporters and editors on how LLMs work, why they hallucinate, and practical techniques like ‘Chain-of-Thought’ prompting which involves asking the AI to explain its reasoning.
Newsrooms can also maximise AI’s efficiency gains while reinforcing the core, irreplaceable value of human journalism. The media house leadership must redeploy human journalists for better purposes. Using AI for grunt work like data summarisation, referencing and data analysis can help efficiency and redirect human journalists to focus on in-depth reporting, unique analysis and original newsgathering—the areas where AI currently cannot compete. This requires newsrooms to prioritising and focus on high-value journalism and discourage its input team from scheduling graphic dependent, repetitive, knee-jerk news items often harvested from third party sources.
The second effort required in the newsroom is to understand the AI’s probabilistic nature which makes human verification non-negotiable for high-impact content. The management needs to implement a ‘Digital Vetting Protocol’. All AI-generated content like summaries, drafts, data points and quotes must be routed through a dedicated fact-checker or senior sub-editor trained specifically in AI-generated errors. This effort calls for making the Human-in-the-Loop in the newsroom mandatory for all the content generated at any point of time.
Another important requirement for mitigation of worries due to AI hallucination is to build public trust by being transparent about the tools in use in the newsroom and strictly adhere to emerging government advisories on synthetic content. Mandatory labelling and watermarking is one way of doing it. It helps implement clear, visible disclaimers on all stories substantially assisted by AI, such as “AI-Assisted Content – Human-Vetted.”
Another popular practice in international media houses, which could help Indian media, is the adoption of a technique which forces the AI to ‘look up’ and anchor its output in verifiable, trusted sources, significantly reducing fabrication. It’s called the Retrieval-Augmented Generation (RAG). The protocol integrates AI tools with the newsroom’s own verified, vetted, and authoritative internal archives and databases. RAG has lately become the most authentic methodology for source grounding in European media houses.
Total elimination of AI hallucination may be impossible given current model architectures. Therefore, the focus must shift from preventing fabrication to enforcing factual grounding. Retrieval-Augmented Generation (RAG) is the most promising technical strategy for achieving this in the newsroom.
International regulatory efforts, particularly in the European Union, are establishing a clear global benchmark for mitigating AI risks, including hallucination, especially where the content impacts public interest. The EU AI Act uses a risk-based approach to regulation and it imposes specific requirements that directly address the hallucination problem in public communication. Though the Act doesn’t classify media as High-Risk, it places systems used in democratic processes or critical infrastructure under the strict High-Risk category. This sets a precedent for how AI used to inform the public on matters of critical importance must be treated. It mandates that providers of General Purpose AI (GPAI) must ensure that any generated content (like text, audio, or video) is clearly identifiable as artificially generated. This requirement focuses on preventing unintentional deception and the spread of ungrounded information. The Act specifically requires that AI-generated content intended to inform the public on matters of public interest must be clearly and visibly labelled. This provision directly targets the risks of hallucination in journalism, compelling disclosure for news-style content.
While not a government regulation, the Associated Press, a major global news agency, was one of the first to implement strict self-regulatory guidelines that serve as an industry standard for avoiding hallucination. The AP guidelines make it mandatory that no generative AI output can be published without human review by an editor who is accountable for the facts, reinforcing the human oversight principle. The AP explicitly bars using generative AI to produce information that purports to be factual or quotes that are attributed to individuals or organizations unless they are used solely as a starting draft that is fully and independently verified by a human reporter. These international efforts reinforce the consensus that to avoid AI hallucination, policy must enforce accountability on the publisher and transparency for the audience.
In the newsroom globally, AI has demonstrated its usefulness in several areas. However, unlike human journalists, it cannot perform physical tasks such as court reporting, interviewing a person, meeting with the families of a tragedy, speaking in confidence with a whistleblower, building faith with the sources or covering a war front. However, there are many tasks for which news organizations can turn to AI like access and observation of facts and figures, selection and filtering of information, processing and editing copies for language errors, publishing and distribution of news material. To conclude, the ghost in the newsroom is a reflection of the industry’s own choice whether to use AI merely for acceleration or to use it to ease the newsroom of cliche jobs and use the journalists to elevate the standard of public-interest journalism. By investing in robust editorial firewalls and AI literacy, Indian media can ensure that AI becomes a powerful tool for truth and not an engine for confident misinformation.
[ Santosh Koshy Joy is a Journalist and Doctoral Scholar with the Social Science Department of the University of Delhi. His research pertains to the impact of AI in learning systems. Opinions expressed are personal. ]
Get real time update about this post category directly on your device, subscribe now.
