Meta's Indian Dilemma: Algorithms, Audits, And Safe Harbour

As Meta enters the second week of talks with the government, the tussle is increasingly centred on a set of questions ranging from how much human intervention should be built into content moderation, and whether reviewers need to understand Indian languages, polity and local context, to how deepfakes should be labelled, how quickly unlawful content must be taken down and whether Meta's recommendation systems could affect its intermediary status, officials familiar with the matter said.While legal experts differ on the scope of the government's powers and Meta's obligations, they broadly agree that the dispute turns on how far intermediary protections extend when platforms actively recommend and amplify content. They question whether automated recommendations can make a platform a publisher, how far the government can mandate human review and algorithmic oversight under existing rules, and whether faster takedown obligations can be reconciled with free speech and due process. They have also flagged concerns over Meta's investment in Indian-language moderation, the legal and constitutional implications of deepfake labelling, and the evolving global approach to platform liability.Here are the five key legal and policy questions at the heart of the discussions, and what experts say about them:1. Can Meta's recommendation algorithms affect its safe-harbour status?At the centre of the government's broader examination is the question of whether Meta continues to meet the requirements for intermediary protection under Indian law.Officials familiar with the matter said the government is examining whether Meta complies with intermediary provisions or whether aspects of the way it operates could affect its eligibility for safe-harbour protection.A key part of the government's argument is that if platforms decide what content is shown to users, that could begin to resemble publishing, officials said. The same question arises when a platform's recommendation system determines what content is shown to whom, or when paid content is promoted. "If platforms decide what content is shown, it is akin to publishing and they must take responsibility for their actions," officials familiar with the matter said.Apar Gupta, founder of the Internet Freedom Foundation, a digital rights advocacy think-tank, however, disagreed with the government's interpretation of recommendation systems.He said Section 79(2)(b) of the Information Technology Act contemplates a human choice in selecting the receiver of a transmission, whereas automated ranking and recommendation systems apply predetermined criteria across millions of posts based on users' activity and preferences. "No. Section 79(2)(b) speaks to selecting the receiver of a transmission. That contemplates a directed, human choice," Gupta said. He argued that treating automated recommendations as equivalent to selecting content could have consequences far beyond Meta. "Here, a contrary reading as incorrectly argued by the government would strip search ranking, spam filters, app stores and streaming recommendations of immunity, and hands government a lever over every ranked service," he said.Gupta also said Indian law does not have a separate "publisher" category for platforms. Section 2(1)(w), he said, defines an intermediary by function, while Section 79 makes immunity conditional on the platform not initiating, selecting or modifying content and on compliance with due diligence requirements. "Indian law has no publisher category for platforms," Gupta said. He said losing Section 79 protection would expose Meta to ordinary liability, rather than automatically transforming it into a publisher. "Ministers calling platforms publishers is a bargaining posture and a political argument," he said.Jyoti Panday, independent researcher on tech-policy, said intermediary liability laws were originally designed to protect technology companies by treating them as passive hosts or "dumb pipes". "Intermediary liability laws include safe harbour provisions which were designed with the intent of protecting tech companies by treating them as passive hosts or dumb pipes," Panday said. In India, she said, safe harbour or immunity is conditional on platforms taking action with regard to specified illegal content categories within a specified period of time under Section 79 of the IT Act."In India the safe harbour or immunity is conditional on platforms taking action with regard to specified illegal content categories within a specified period of time," Panday said.Panday said the issue is also part of a wider global shift in intermediary-liability law. "Globally intermediary liability regimes are rapidly evolving," she said.The key question, she said, is whether a platform's recommendation system remains merely a mechanism for hosting content or crosses into active curation, prioritisation and amplification of unlawful material "When a platform's recommendation algorithm moves past hosting and actively curates, prioritises and amplifies illegal content then platforms could face losing their immunity," Panday said.Panday pointed to the US litigation involving TikTok as an example of how courts are beginning to examine recommendation algorithms separately from the underlying third-party content. "For eg in Anderson vs TikTok, the 3rd circuit held that TikTok's algorithmic recommendations constitute as its own expression," Panday said. She noted that the US legal framework is different from India's. In the US, Section 230 of the Communications Decency Act provides immunity for third-party content, but, she said, that protection does not necessarily extend to claims based on a platform's own recommendation or expression. "Since section 230 of CDA in the US provides immunity for 3rd party [content] it does not protect TikTok from liability for its own recommendation," Panday said.The distinction is important to the government's examination of Meta because officials are looking not only at what users post but also at what the platform's systems do with that content. Officials said the government is examining whether recommendation engines and paid-content systems create responsibilities beyond those traditionally associated with an intermediary.The ultimate determination of whether a platform is an intermediary or publisher would be for the courts to make, officials said.2. Can the government mandate human review and should reviewers understand Indian languages and politics?The strongest operational demand emerging from the talks is for greater human intervention in content moderation, officials said. The government believes automated moderation can struggle to distinguish between content that is genuinely harmful or unlawful and material that may be political satire, commentary or legitimate speech when viewed in the Indian context.Officials said the government has therefore sought stronger human checks, with reviewers who understand Indian languages, India's polity and local political and social context. The demand has particular significance for sensitive cases involving prominent public figures. Officials said the government has sought a system involving human intervention before automated action is taken in such cases, with at least two people involved in reviewing content involving prominent personalities."People familiar with Indian languages, politics and context are necessary because automated systems may not understand nuances that are obvious to a human reviewer familiar with the country," a senior government official said.The question also ties into the broader legal debate over whether platforms are merely passive hosts or are making substantive decisions about content.Panday's analysis of intermediary liability places that distinction at the centre of the debate: safe harbour was designed around the idea of platforms acting as passive hosts, while concerns arise when their systems actively curate, prioritise and amplify content.Gupta said the government can mandate some human oversight under the existing IT Rules, but that its ability to go much further is legally questionable."Yes, but only partially," Gupta said, arguing that the Union government is attempting to push several regulatory directions through intermediary-liability provisions in the absence of a dedicated platform-regulation law. He pointed to Rule 4(4) of the IT Rules, which requires significant social media intermediaries to deploy automated tools for certain categories of content involving rape and child abuse imagery, with human oversight and review by the platform. Rule 4(1)(a), he said, also requires a resident Chief Compliance Officer.But Gupta argued the existing framework does not mandate human review of every automated moderation decision. "Nothing mandates human review of each automated decision. Nothing empowers MeitY to demand ranking weights or source code," he said.He also argued that Section 87 permits due-diligence guidelines through delegated legislation but does not necessarily give the government a general audit power over platform algorithms.The government's concern, officials said, is that human intervention must be meaningful and capable of understanding the Indian context rather than merely being a procedural check after an automated system has already made a decision. That concern was brought into sharper focus by the government's objection to action taken against a post involving Prime Minister Narendra Modi.3. What legal duties does Meta have when content violates Indian law but not its own policy?The government has raised concerns about content that officials say does not necessarily violate Meta's community guidelines but may nevertheless raise concerns under Indian law. Child sexual abuse material (CSAM) content with public-order implications and other potentially unlawful material are among the issues raised with Meta, officials said. The government's position is that a platform's community standards cannot by themselves determine whether content complies with Indian law.Officials said the company has been told that it operates in India and must comply with Indian law even where a particular piece of content does not breach its own community standards.At the same time, officials acknowledged that not every piece of content flagged by authorities is necessarily unlawful. In one set of cases involving content mentioned in an FIR filed in Hyderabad, the majority of the material was blocked pursuant to legal orders, officials said. But some of the content did not violate Meta's policies. Several users were booked over allegedly morphed photographs. Officials said some of the material involved deepfakes, while other cases involved political satire, which can be protected under Meta's free-speech policies.Meta, on the other hand, released a statement that it was in touch with the relevant authorities. According to people familiar with the matter, most of the content that was flagged was taken down, but posts that could constitute free-speech or political satire were not taken down.The distinction has added to the government's emphasis on human review and contextual understanding. Gupta said Meta has obligations of transparency and accountability towards its Indian users beyond what is contained in the IT Rules. "Meta over and above the poorly considered IT Rules, 2021 has a duty of transparency and accountability to its users in India," he said.He also pointed to concerns over the company's investment in Indian-language content review, citing disclosures that he said showed a significant imbalance in resources devoted to English-language misinformation classification. "Meta also under-invests in Indian language review," Gupta said. He cited the Haugen disclosures as showing that 87% of Meta's misinformation-classification budget covered English.But Gupta also cautioned against treating platform companies as the sole actors responsible for the online information environment. "We need to remember a basic civics lesson that while we hold Silicon Valley companies responsible for their wrongs, these wrongs are tolerated and permitted by an elected government which acts through the Ministry of Electronics and IT," he said.The government, meanwhile, has said the issue is not simply about Meta's internal policies but about the company's obligations under Indian law. The debate over safe harbour also feeds into this question. Panday said the protection is conditional on platforms acting on specified categories of illegal content within prescribed periods, making compliance with takedown requirements a central part of the intermediary framework.4. How should deepfakes be labelled and can the government demand stronger safeguards?Deepfakes remain a major area of concern for the government. Officials familiar with the matter said the government is fundamentally concerned about Meta's approach to labelling manipulated and synthetic content. They said compliance with the government's expectations has been partial and that Meta has indicated that it is prioritising prominent personalities. The government believes the problem cannot be limited to a small group of high-profile individuals and has asked Meta to strengthen its approach to deepfakes more broadly, the officials said. The government will review the outcome of the meetings before taking a view on further action.Deepfakes have become a particularly sensitive issue because manipulated material can spread rapidly and potentially affect public order, political discourse and public trust, they said. The government is examining both the technical and legal dimensions of the issue, including whether existing safeguards are sufficient and whether independent audits of algorithms should be considered.Gupta said Meta already has labelling mechanisms on its platforms and pointed to the February 2026 amendments to the IT Rules, which introduced requirements relating to synthetically generated information. "Labels are already present on META platforms," Gupta said. He said the amended rules, notified on February 10 and effective from February 20, mandate prominent labelling of synthetically generated information with permanent provenance metadata and require significant platforms to verify certain user declarations before publication.Gupta said the Internet Freedom Foundation had raised concerns about compelled speech and the reliability of detection tools. "IFF's comments on these rules flagged compelled speech and unreliable detection tools to warn of further unconstitutional censorship that will occur under the guise of a social panic around deepfakes," he said.The government is also examining whether independent audits of algorithms could form part of a stronger checks-and-balances framework. Officials said technical teams from the two sides will continue discussing Meta's safeguards, including the technical guardrails around automated moderation and the process for identifying and labelling manipulated content.5. Can the government require three-hour takedowns while preserving due process and free speech?The government has reiterated its position on faster takedowns, including the three-hour timeline introduced through the February amendment to the IT Rules. Officials said the government believes faster action is necessary because unlawful content, particularly manipulated or synthetic material, can spread rapidly across platforms. The three-hour requirement has become another point of discussion with Meta as the government weighs the need for swift enforcement against the need for human review and contextual assessment.After its second round of meetings with Meta, Meity officials said the platform has committed to greater due diligence when it comes to addressing unlawful content.The government's position is that platforms have the technology and resources to comply with the shorter timeline and that speed is increasingly important where unlawful content can spread rapidly. The issue nevertheless raises questions about how platforms can make legally and contextually sensitive decisions within such a short period. That tension runs through the government's discussions with Meta: faster action is being sought at the same time as greater human intervention and India-specific contextual expertise.The government has reiterated that the three-hour requirement applies to specified unlawful-content takedown directions under the amended framework, officials said.Panday's assessment of intermediary liability places the speed requirement within the larger safe-harbour bargain: platforms receive immunity from liability subject to complying with specified obligations, including action against specified categories of illegal content within prescribed timelines.Gupta said the government's regulatory approach is attempting to address a wide range of platform issues through intermediary-liability provisions, raising questions about the limits of delegated regulation.The three-hour requirement itself is now part of the amended rules, however, and the government has reiterated that it expects platforms to comply.The demand for greater human intervention gained urgency after the government raised concerns over action taken against content involving Prime Minister Narendra Modi. The episode has reinforced the government's concern that algorithms may not always understand the significance or context of political content in India."I apologised to the minister on behalf of Meta for the error restricting PM Modi's post," Joel Kaplan, chief global affairs officer at Meta, said.Exploring a larger frameworkThe government is not limiting its examination to Meta.Officials said other platforms may also be called in as the government examines whether they meet the definition of intermediaries under Indian law and whether their operations comply with the obligations attached to that status. Officials described the issue as extending beyond a dispute between the government and one technology company.The government believes global platforms operating in India must respect the country's laws and constitutional framework, while also ensuring that enforcement does not unnecessarily restrict legitimate speech. "Rule of law and the Constitution" are central to the government's approach, officials said, adding that maintaining trust in digital platforms is critical.With the global leadership-level meetings already held twice, the next phase of engagement will be handled by Meta's India and technical teams, officials said. Those teams will continue discussions on moderation systems, human intervention, deepfake safeguards and the technical guardrails around automated decisions.