AI in the newsroom and playout—Automation without diluting editorial control and compliance

Artificial Intelligence is now embedded across newsroom operations, shaping content discovery, production, packaging, and distribution strategies, and is increasingly becoming part of the editorial infrastructure. The upside is operational speed, cost efficiency, and scale, while the downside is structural, diluted editorial control, heightened regulatory risk, and a potential hit to credibility. Global news organizations have integrated AI systems into workflows for transcription, translation, archive search, and summarization. On the broadcast side, synthetic anchors and AI-assisted playout have already been deployed. At this juncture, the primary challenge is disciplined deployment of AI. Automation must be calibrated so that it augments editorial judgment without displacing it and accelerates distribution without compromising compliance.

Upstream movement of automation into editorial judgment
Early deployments of AI systems in newsrooms were confined to low-risk, predictable tasks that are mostly data-driven and easy to validate. That boundary, however, has now shifted, and AI systems are now being used in, inter alia:

• Story discovery through trend and signal detection,
• Summarisation of large document sets in investigative reporting,
• Draft generation for general news copy,
• Headline and engagement optimization, and
• Multilingual translation and localization.

Decisions regarding what gets surfaced, how it is framed, and how widely it is distributed carry editorial weight. The practical consequence of AI deployment is that editorial judgment is increasingly exercised at the level of system design, including training data, prompt structures, and ranking logic, rather than at the point of publication. That redistribution is operationally efficient, but legally significant, because accountability frameworks have not moved.

The convergence of regulatory frameworks
The regulatory response to AI in media is three consistent expectations: accountability, transparency, and human oversight.

Under the Digital Personal Data Protection Act, 2023, the use of personal data within AI systems, whether for recommendation, profiling, targeting, or optimization, must be anchored in a lawful basis and consent, and be confined to specified purposes. Obligations relating to purpose limitation, data minimization, accuracy, and reasonable security safeguards apply equally to automated processing. When news publishers rely on behavioral data to personalize or prioritize content, they act as data fiduciaries and must be able to justify both the collection and the downstream use of such data for clearly defined purposes. The opacity of AI-driven decision-making does not dilute these requirements. If anything, it raises the threshold for demonstrable compliance, particularly where profiling materially influences content exposure, with breaches attracting monetary penalties of up to INR 250 crore for failure to implement reasonable security safeguards, and up to INR 200 crore for breaches relating to children’s data or other specified obligations under the Act, in addition to advertiser withdrawal, platform sanctions, and erosion of audience trust.

This position is also embedded in the structure of the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021. The 2023 amendments enhance obligations on intermediaries and digital publishers to prevent the dissemination of “false, fake or misleading” information, including content flagged through government-notified fact-checking processes, while the accompanying Code of Ethics imposes standards of accuracy and non-distortion. Read together, these provisions shift the emphasis from reactive takedown to upstream responsibility in content creation and dissemination. This is reinforced by the March 2024 advisories issued by the Ministry of Electronics and Information Technology on AI systems, which emphasise due diligence, testing, and the prevention of bias and unlawful outputs, with non-compliance risking loss of safe harbour protection under Section 79 of the Information Technology Act, 2000, thereby exposing publishers and intermediaries to direct liability for third-party content, alongside blocking orders under Section 69A.

Within the broadcast ecosystem, obligations under the Cable Television Networks (Regulation) Act, 1995, and the Program Code framed under Rule 6 of the Cable Television Networks Rules, 1994, continue to apply. These include, in particular, prohibitions on content that is false, suggestive of innuendo or half-truths, content that offends decency or morality, and content likely to incite harm or disturb public order. These standards are indifferent to the mode of production. Whether a segment is scripted by a journalist, translated by an automated system, or delivered through a synthetic voice layer, the broadcaster remains responsible for ensuring that the output meets these substantive requirements. Violations thereof may lead to program takedown, prohibition of transmission, suspension of channel permissions, and imprisonment of up to 2 years and/or fines.

The regulatory concern, therefore, extends to routine failures inherent in AI-assisted workflows, mistranslation that alters meaning, synthetic presentation that implies authenticity, or automated clipping that removes context and distorts narrative. Increasingly, the focus is shifting from isolated instances of non-compliant content to the adequacy of the systems that generate and disseminate it. This includes whether validation layers exist, whether higher-risk outputs are subject to human review, and whether there are real-time mechanisms to halt or correct unreliable content. There is no emerging safe harbor for artificial intelligence systems. If anything, their scale, speed, and opacity are prompting closer scrutiny, with failures likely to be characterized not as editorial lapses, but as deficiencies in system design and governance.

Liability and attribution in AI-assisted editorial workflows
The central legal issue is not the applicability of existing law, but the attribution of responsibility within fragmented, AI-assisted workflows. Traditional editorial chains, writing, review, and publication are replaced by systems where content is generated, minimally edited, and distributed without a clear point of control. Despite this, liability remains anchored in publication. The entity that disseminates and monetizes the content continues to bear responsibility, with no dilution of responsibility due to automation.

The inquiry, therefore, ranges from isolated editorial decisions to whether the system itself was designed and operated to meet legal standards. Crucially, this distinction maps onto the allocation of responsibility between a data fiduciary and a data processor. The newsroom or broadcaster, as the entity determining the purpose and means of processing, remains the data fiduciary and bears primary legal responsibility. Technology vendors and AI providers serve as processors that execute instructions. As a result, even where systems are externally sourced, the obligation to ensure lawful processing, accuracy, and compliant dissemination remains with the deploying organization. In effect, accountability extends beyond editorial teams to include those responsible for system design, deployment, and governance, while remaining anchored in the fiduciary who controls the system’s use.

Conclusion and next step
Primary responsibility continues to rest with the publisher or broadcaster, regardless of how content is generated or distributed. Where systems influence editorial output, the organization must be able to explain, and stand behind, both the outcome and the process that produced it.

In practice, this calls for a more deliberate approach than most current deployments reflect. AI use needs to be mapped to specific functions, with clear SOPs setting out when human intervention is required, particularly where content carries legal, societal, or market sensitivity. Equally, organizations need to be in a position to reconstruct how content was produced, what data was used, what the system generated, and where editorial judgment was applied.

For platforms and media organizations, this requires a shift from policy-level commitments to enforceable operational controls. At a minimum, organizations should: (i) clearly classify AI use-cases by risk and mandate human review for high-impact outputs, particularly in political, financial, or real-time broadcast contexts; (ii) maintain audit trails for AI-generated or AI-assisted content, including inputs, outputs, and intervention points, to meet evidentiary and regulatory expectations; (iii) ensure DPDP compliance by mapping all AI-driven personal data use to specific purposes backed by valid consent, with demonstrable adherence to data minimization and security safeguards; (iv) build system-level controls in playout environments, including confidence thresholds, escalation triggers, and override mechanisms; and (v) restructure vendor contracts to address training data provenance, audit rights, and liability allocation, rather than treating AI tools as standard software procurement.

At its core, the issue is not whether AI is used but how well its deployment is controlled. The legal standard has not moved; only the way content is produced has. That means the burden on organizations is higher. If systems are doing part of the editorial work, they need to be governed like editorial functions. If they are shaping output, they need to be auditable. And if something goes wrong, the organization must be able to explain why. Where that discipline exists, AI is manageable. Where it does not, the risk is not theoretical; it shows up the moment something slips through.

The article is co-authored by Karnika Vallabh (Counsel), and Nandini Tyagi (Associate), Bharucha & Partners